JP2001117922A - Translation apparatus, translation method, and recording medium - Google Patents

Abstract

(57) [Summary] [Problem] To perform high-accuracy translation and to output a translated sentence with a synthesized sound that does not cause discomfort. SOLUTION: A speech recognition unit 1 recognizes an input speech, extracts prosody information of the input speech, and supplies it to a machine translation unit 2 together with a speech recognition result of the input speech. The machine translation unit 2 has a conversion table in which language conversion data for performing Japanese-English and English-Japanese translation is described together with prosody information on English and Japanese, and based on the conversion table, The speech recognition result is translated into a translation with prosody information. The translated sentence with the prosody information is supplied to the speech synthesizing unit 3, and is output as a synthesized sound from the speaker 5 by performing a regular speech synthesizing process.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a translation apparatus, a translation method, and a recording medium, and more particularly to a translation apparatus that performs high-precision translation by performing translation in consideration of prosody information. This enables a sentence to be output as a highly accurate synthesized sound.

[0002]

2. Description of the Related Art A speech translation system is a tool for communicating between users who speak in different languages such as Japanese and English. In a speech translation system, for example, a Japanese utterance is speech-recognized, the speech recognition result is translated into English and output as a synthesized sound, and the English utterance is speech-recognized. And output as synthesized speech. Therefore, an English speaker (user) can hear the utterance of a Japanese speaker in English, and a Japanese speaker can hear the utterance of an English speaker in Japanese. They can interact with each other by understanding the utterance of the other party.

FIG. 1 shows the configuration of an example of a conventional speech translation system.

[0004] Speech in a source language (language before translation) is input to a speech recognition unit 201, where the speech is recognized. The speech recognition unit 201 outputs the speech recognition result of the speech in the source language to the machine translation unit 202 as a text. In addition,
1 from the speech recognition unit 201 to the machine translation unit 202
A plurality of voice recognition result candidates may be output instead of one voice recognition result.

[0005] In the machine translation unit 202, the text of the speech recognition result in the source language is translated into a text (translation sentence) in the target language (the translated language) and supplied to the speech synthesis unit 203. The speech synthesis unit 203 generates the phoneme information based on the translated sentence from the machine translation unit 202, and generates and outputs a synthesized speech corresponding to the translated sentence by adding an appropriate intonation. The translated sentence output by the machine translation unit 202 may be displayed on a display (not shown).

[0006]

In the speech translation system, regarding the processing accuracy of whether a speech in the source language is output as a synthesized speech corresponding to a correct translation in the target language, the speech recognition unit 201 and the machine translation The accuracy of the processing in each of the unit 202 and the voice synthesizing unit 203 becomes a problem. In addition, processing in which voice (voice in the original language) is temporarily converted into text and then converted again into voice (synthesized voice in the target language) Missing information during the process can also be a problem.

That is, speech includes not only phonological information that can be represented by readable characters but also prosody information on prosody such as accent, intonation, rhythm, and pause. The recognition is performed, and the speech recognition result is output as text, so that the prosody information included in the speech is lost.

[0008] However, even if the voice is the same when expressed by characters, the meaning may differ depending on the prosody information included in the voice, and as described above, the text is converted during the translation process. The loss of prosody information due to intervening may affect the translation result.

That is, for example, in order for Japanese and English speakers (users) to communicate with each other, it is assumed that translation between Japanese and English and between English and Japanese is performed. *) Thus, for example, "* I * have a pe
n. "means pronounce" I "strongly and" I havea
* pen *. "indicates that" pen "is pronounced strongly.

For example, in a sentence, a phrase which is new information (information newly appearing in a dialogue) and old information (a word in a dialogue,
Information that has already appeared) and the true information appears before the old information,
In English, new information phrases are generally uttered with accents. Therefore, if the translation is performed without considering the accent, a strange translation result may be obtained (an incorrect translation result may be obtained).

Specifically, for example, when there is a fact that "my brother took a picture", the Japanese utterance "this picture was taken by my brother" It is information, and "brother" is new information. The English translation (translation) of this Japanese utterance is "My
* brother * took this picture. "
A translation of "My brother" is translated into a translation of the old information "This photo".
s picture "appears before the English translation" My * broth
In "er * took this picture.", "brother" is uttered with an accent.

On the other hand, as for the utterance in Japanese, "My brother took this picture", "This picture" is new information and "Brother" is old information. The English translation of this Japanese utterance is "My brother took this pictur
e. "and the new information" this picture "
However, the English translation of "My brother took this picture."
Spoken without accents.

The English translations of the utterances in Japanese, "This picture was taken by my brother." And "My brother took this picture."
s picture. ", which is the same, but the accent when speaking is different as described above. Therefore, when translating Japanese to English or English to Japanese without considering the accent, the following problem occurs. Occurs.

That is, when a Japanese-English translation is performed and the translation result is output as a synthesized sound, the English translation of "My * brother * took t" in Japanese utterance "This picture was taken by a younger brother."
The synthesized sound of "his picture." should have an accent on "brother", and the English translation of "My brother took this pi" in Japanese utterance "My brother took this picture."
cture. "should have no accent.

However, when the accent is not considered, any English translation is output as a synthesized sound without accent. In other words, the translation results of the utterances in Japanese, "This picture was taken by my brother." And "My brother took this picture."
ok this picture. "Therefore, for English users, the English translation of the Japanese utterance," This picture was taken by my brother. " , A synthetic sound with an uncomfortable feeling is output.

Conversely, when translating English to Japanese, the English utterance "My * brother * took this picture."
As described above, the translation result of "other took this picture." is, "This picture was taken by my brother."
"My brother took this picture." However, if accents are not taken into account, the above two utterances in English are indistinguishable, and both Japanese translations say, "My brother took this picture."
And for Japanese users, the utterance "M
y * brother * took this picture. "
A translated sentence with a sense of incongruity is output.

For example, in Japanese, the particles "wa" and "ga" may be used properly, and for the particle "ga", the noun placed immediately before it is uttered with an accent. However, if the translation is performed without considering the accent, an unnatural translation result may be obtained.

Specifically, for example, a user's response in English to the question "What happened to Henry?" Is "Henry has arrived." Or "Henry has * arrived."
*. ", The Japanese translation is" Henry has arrived. "

For example, when the user's reply in English to the question "Who has arrived" or "Who has arrived" is "* Henry * has arrived." The Japanese translation is "* Henry has arrived *" or "Henry has arrived."

Further, for example, a user's response in English to the question "what happened" is "Henry
In the case of "has arrived.", the Japanese translation is "Henry has arrived."

Therefore, in the Japanese-English translations, "Henry has arrived,""Henry has arrived,""Henry has arrived," and "Henry has arrived." The translation results are all "Henry has arrived." If the accent is not taken into account, and if this translation result is output as a synthetic sound, a synthetic sound that is semantically strange for an English user is output. May be done.

On the other hand, in the English-Japanese translation, "Henry has
arrived. "," * Henry * has arrived. "," Henry has * ar
The translation of "rived *.", without considering accents, is either a response to the question "What happened to Henry?" or "Henry has arrived.""Henry has arrived." Is the answer to the question "That?", And the translation cannot be performed due to the difference in accent in the English utterance. As a result, a translated sentence with a sense of incongruity may be output for a Japanese user.

Further, for example, in English, an utterance with an accent may be performed depending on a positional relationship between an adverb and a semantic destination of the adverb. Translation results may be obtained.

That is, in English, for example, "also" or "only"
Are basically placed just before the word it modifies ("only" may be right after the word it modifies), but in spoken language these adverbs are Is placed immediately before a verb, and an accent (stress) may be added to a word of a modification destination.

For this reason, for example, the utterance in English "I als
The Japanese translation of "o like * her *." should be "I also like her."
The Japanese translation of "ke her." should be "I also like her."

Further, for example, the utterance in English "I only sa
The Japanese translation of "w * her * yesterday." should be "I met her only yesterday." or "She was the only person I met yesterday.""I o
The Japanese translation of "nly saw her * yesterday *." should be "I met her only yesterday." or "I was only with her yesterday."

Further, for example, in English, when several phrases are connected to form a compound word, the accent moves forward.

For this reason, for example, "English teacher"
The noun phrase "(an) English * teacher *" should be translated into English, and the compound noun phrase "English teacher" should be translated into "(an) * English * teacher". is there.

Therefore, if the translation is performed without considering the accent, as in the above-described case, in the Japanese-English translation, the English translation is always output as an accent-free synthesized sound, and as a result, for the English user, In some cases, a synthetic sound having a sense of incongruity may be output. On the other hand, even in English-Japanese translation, translation cannot be performed due to differences in accents in utterances in English, and as a result, a translated sentence with a sense of incongruity may be output to a Japanese user.

In addition, in general, even if the utterance format is a declarative sentence, when the intonation at the end of the sentence is raised (assuming the intonation of the rising tone), the sentence can be a sentence intended for a question. Therefore, if intonation is not considered, even if the utterance is intended to be a question, the utterance will be
All are translated as declarative sentences, and the translation results are output as synthesized sounds. As a result, a translation that does not meet the user's intention is output.

Furthermore, from the viewpoint of the man-machine interface of the speech translation system, it may be desirable to output a synthesized speech as a translation result in accordance with the personality of the user. That is, as the personality of the user, for example, the pitch of the voice (for example, whether the voice is male or female), the speed of speech (for example, whether the voice is fast or slow), age, tone ( (Eg, angry or happy), speech power, and other non-verbal sound information (eg, laughter, sneezing, tongue)
However, when a translation of a user's utterance is output using a synthetic sound that reflects the user's personality, it is possible to more accurately convey his or her intentions to the other party of the dialogue. Become.

However, in the conventional speech translation system, once the utterance of the user is converted into text, the prosody information reflecting the personality of the user as described above is lost. It is difficult to output the reflected synthesized sound.

For example, when it is desired to change the translation result of the utterance by the user according to the gender of the user, the gender can be identified relatively easily from the voice. However, it is difficult to perform the identification after the text is converted into the text as the speech recognition result.

Therefore, some of the conventional speech recognition and speech translation techniques perform processing using speech prosody information.

That is, for example, Japanese Unexamined Patent Publication No. Hei 8-50498 uses not only phonological information but also intonation information (phonological information), for example, homophonic accents such as "bridge" and "chopsticks". A method for performing speech recognition by distinguishing words is disclosed.

However, in the method disclosed in Japanese Patent Laid-Open No. 50498/1996, words having the same word string but different accented words (for example, the above-mentioned "(an) English * teacher *" , "(An) * English * teac
her "). Furthermore, the method disclosed in Japanese Patent Application Laid-Open No. H8-50498 can only perform speech recognition by distinguishing homophonic accent words, When translating the text output as a result of recognition and generating a synthesized speech corresponding to the translation result, the prosody information is not taken into account. Output cannot be dealt with.

For example, Japanese Patent Application Laid-Open No. Hei 6-332494 discloses a translator that extracts an accented phrase from an input speech in a source language and adds an accent to a target language phrase corresponding to the extracted phrase. ing.

However, in Japanese Patent Application Laid-Open No. Hei 6-332494, when a phrase in the target language corresponding to a phrase with an accent in the source language has an accent (for example, the above-mentioned "* Henry * has arrived." Henry has arrived. "Or" Henry has arriv
ed. "is translated as" Henry has arrived. "However, if the target language phrase that corresponds to the accented phrase in the source language has no accent (for example, "Henry has * arrive"
d *. "is translated as" Henry has arrived. "), and it is difficult to deal with accents in the target language that correspond to unaccented phrases in the source language. .

The present invention has been made in view of such a situation, and is intended to perform high-accuracy translation and to output a translated sentence with a synthesized sound without a sense of incongruity. .

[0040]

According to the translation apparatus of the present invention, an input means for inputting an input sentence and a correspondence relationship for translating the input sentence into a translated sentence are set in a first or second language. Based on a table described together with at least one piece of prosody information, a translation unit that translates an input sentence into a translation corresponding to the input sentence, and an output unit that outputs a translation is provided.

The input means is provided with a voice recognition means for recognizing voice, and the result of voice recognition by the voice recognition means can be output as an input sentence.

The speech recognition means is provided with an extraction means for extracting speech prosody information, and the speech recognition result of the speech can be output together with the speech prosody information. The translation means outputs the speech recognition result. It can be translated using the prosody information.

The translation apparatus according to the present invention may further include request means for requesting prosody information to the voice recognition means. In this case, the voice recognition means may receive a request from the request means. , Can output prosody information.

The output means is provided with a speech synthesizing means for generating a synthesized speech corresponding to the translated sentence, and can output the translated sentence as a synthesized speech.

The translation means can output the translated sentence together with the prosody information, and the speech synthesizing means can generate a synthesized sound corresponding to the translated sentence using the prosody information.

The speech synthesizing means can be provided with a requesting means for requesting the prosody information from the translating means.
In this case, the translating means can output prosody information when requested by the request means.

The translation apparatus according to the present invention stores a table in which a correspondence for translating an input sentence into a translated sentence is described together with prosody information of at least one of the first and second languages. Storage means can be further provided.

[0048] In the translation method of the present invention, the input step of inputting an input sentence and the correspondence relation for translating the input sentence into a translated sentence include the prosody information of at least one of the first and second languages. It is characterized by including a translation step of translating an input sentence into a translation corresponding to the input sentence based on the described table, and an output step of outputting the translated sentence.

According to the recording medium of the present invention, an input step of inputting an input sentence and a correspondence relation for translating the input sentence into a translated sentence are provided together with prosody information of at least one of the first and second languages. A program is recorded which includes a translation step of translating an input sentence into a translation corresponding to the input sentence based on the described table, and an output step of outputting the translated sentence.

In the translation apparatus, the translation method, and the recording medium of the present invention, the correspondence for translating the input sentence into the translated sentence is described together with the prosody information of at least one of the first and second languages. The input sentence is translated into a translated sentence corresponding to the input sentence based on the table that has been set.

[0051]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 2 shows a speech translation system to which the present invention is applied (a system means a plurality of devices logically assembled, and whether each device is in the same housing or not). FIG. 3 shows an example of the external configuration of the speech translation system.

In this speech translation system, when a speech in a source language such as Japanese or English is input, a translated sentence obtained by translating the speech into a target language such as English or Japanese is output. I have.

That is, for example, a voice in Japanese or English is input to the microphone 11 and supplied to the voice recognition unit 1. The voice recognition unit 1 recognizes voice from the microphone 11 and outputs text as a result of the voice recognition and other accompanying information to the machine translation unit 2, the display unit 4, and the like.

The machine translation unit 2 analyzes the speech recognition result output from the speech recognition unit 1 and machine translates the language (original language) of the input speech into a target language such as English or Japanese. The text as a translation result and other accompanying information are output to the speech synthesizer 3, the display 4, and the like. The speech synthesis unit 3 performs a speech synthesis process based on the output of the machine translation unit 2 and the like, and outputs a synthesized sound as a result of translating the input speech into another language.

The display unit 4 is composed of, for example, a liquid crystal display or the like, and displays the result of speech recognition by the speech recognition unit 1, the result of machine translation by the machine translation unit 2, and the like.

The operation unit 6 includes, for example, a cursor key 6A operated when moving a cursor, an enter key 6B operated when confirming a selection, and a cancel operation operated when canceling a selection. An operation signal corresponding to an operation of the operation unit 6 is constituted by a key 6C, and is supplied to the control unit 7. The control unit 7 performs various processes according to operation signals from the operation unit 6. Note that the operation unit 6 may be used in addition to the case described above.
It can be used for inputting characters and performing kana-kanji conversion.

In the speech translation system configured as described above, when speech in the source language is input, the speech is recognized by the speech recognition unit 1 and supplied to the machine translation unit 2. In the machine translation unit 2, the speech recognition result by the speech recognition unit 1 is machine-translated into a target language and supplied to the speech synthesis unit 3. The speech synthesis unit 3 generates and outputs a synthesized speech corresponding to the translation result from the machine translation unit 2.

Next, FIG. 4 shows an example of the configuration of the speech recognition section 1 of FIG.

The utterance of the source language by the user is transmitted to the microphone 11
The microphone 11 converts the utterance into an audio signal as an electric signal. This audio signal is AD (A
(nalog Digital) converter 12. AD converter 1
In 2, the audio signal which is an analog signal from the microphone 11 is sampled and quantized, and is converted into audio data which is a digital signal. This audio data is supplied to the feature extraction unit 13 and the buffer unit 14.

The feature extracting unit 13 converts the audio data from the AD converting unit 12 into appropriate frames, for example,
A feature parameter such as a spectrum, a linear prediction coefficient, a cepstrum coefficient, and a line spectrum pair is extracted, and the
4 and the matching unit 15.

The matching unit 15 is based on the feature parameters from the feature extracting unit 13 and
6, dictionary database 17, and grammar database 1
The voice input to the microphone 11 (input voice) is recognized while referring to 8 as necessary.

That is, the acoustic model database 16 stores acoustic models representing acoustic features such as individual phonemes and syllables in the language of the speech to be recognized. Here, as the acoustic model, for example, HMM (Hidden Mark
ov Model) can be used. The dictionary database 17 stores a word dictionary in which information on pronunciation for each word (phrase) to be recognized is described.
The grammar database 18 stores grammar rules describing how each word registered in the word dictionary of the dictionary database 17 is linked (connected). Here, as the grammar rule, for example, a context-free grammar (CF
G) or a rule based on statistical word chain probability (N-gram).

The matching section 15 stores the dictionary database 1
By connecting the acoustic models stored in the acoustic model database 16 by referring to the 7th word dictionary, an acoustic model (word model) of the word is formed. Further, the matching unit 15 connects several word models by referring to the grammar rules stored in the grammar database 18, and uses the thus connected word models based on the feature parameters, for example, , The HMM method or the like is used to recognize the voice input to the microphone 11.

The result of speech recognition by the matching unit 15 is output, for example, as a text in the source language.

Here, in addition to the general-purpose acoustic model, the acoustic model database 16 may include, for example, a male voice, a female voice, and a non-verbal voice (for example, laughter, sneeze, etc.). , Tongue tapping, etc.) can also be stored. In this case, in the matching unit 15, not only a general-purpose acoustic model,
By performing speech recognition also using a special acoustic model, based on the likelihood of the speech recognition result by each acoustic model, whether the input speech is a male utterance or a female utterance, It is possible to determine whether tongue tapping, laughter, or the like.

It should be noted that the matching unit 15 normally performs speech recognition using only a general-purpose acoustic model, and performs speech recognition using a special acoustic model from the machine translation unit 2 and the speech synthesis unit 3. Only when a request signal requesting the above is received, it is possible to perform speech recognition using a special acoustic model as well.

On the other hand, the buffer unit 14 comprises an audio data buffer 14A and a feature buffer 14B. The audio data buffer 14A stores the audio data output from the AD converter 12, and the feature buffer 14B stores the feature extraction unit 1
3 is temporarily stored.

The audio data stored in the audio data buffer 14A and the characteristic parameters stored in the characteristic amount buffer 14B are read by the prosody information extracting unit 19 as necessary, and
9 extracts prosody information of the input voice (input voice) using the voice data and the feature parameters, and converts prosody data representing the prosody information into a text as a voice recognition result output by the matching unit 15. Output as accompanying information.

The prosody information extracting unit 19 can always extract prosody information and output prosody data, and requests the prosody information from the machine translation unit 2 or the speech synthesis unit 3. It is also possible to extract prosody information only when a request signal is received.

The prosody data output from the prosody information extracting unit 19 can be output to the matching unit 15. In this case, the matching unit 15
In the text as the speech recognition result, data including the prosody data from the prosody information extracting unit 19 is generated and output.

FIG. 5 shows an example of the configuration of the machine translation unit 2 shown in FIG.

A text as a result of speech recognition output from the speech recognition unit 1 is input to the text analysis unit 21 as a target of machine translation. The text is analyzed with reference to the usage grammar database 25.

That is, the dictionary database 24 stores a word dictionary in which notation of each word (word) and part of speech information necessary for applying the grammar for analysis are described.
The analysis grammar database 25 stores analysis grammar rules in which restrictions on word chains are described based on information of each word described in the word dictionary. Then, the text analysis unit 21 performs morphological analysis, syntax analysis, and the like of the text (input text) input thereto based on the word dictionary and the grammatical rules for analysis, and executes the words and syntaxes constituting the input text. Linguistic information such as the information of Here, examples of the analysis method in the text analysis unit 21 include a regular grammar, a context-free grammar, and a method using a statistical word chain probability.

The linguistic information as an analysis result of the input text obtained by the text analyzing unit 21 is supplied to the linguistic conversion unit 22. The language conversion unit 22 includes a language conversion database 2
6, the language information of the language (source language) of the input text is converted into the language information of the language (target language) of the translation result.

That is, the language conversion database 26 stores a conversion pattern from the language information of the original language (the language of the input to the language conversion unit 22) to the language information of the target language (the language of the output from the language conversion unit 22). Correspondence between source language and target language for converting linguistic information, such as a translation example between the source language and target language and a thesaurus used to calculate the similarity between the translation example and the source language Is stored. Then, the language conversion unit 22 converts the language information of the language of the input text into language information of the target language based on such language conversion data.

The language information of the target language obtained by the language conversion unit 22 is supplied to a text generation unit 23. The text generation unit 23 refers to the dictionary database 27 and the generation grammar database 28, and
From the language information of the target language from, a text in which the input text is translated into the target language is generated.

That is, the dictionary database 27 stores a word dictionary in which information such as the part of speech and inflected forms of words (phrases) necessary to generate a sentence in the target language is stored. The grammar database 28 stores grammar rules for generation, such as word utilization rules and word order restrictions, necessary for generating sentences in the target language. Then, the text generator 23 converts the linguistic information from the language converter 22 into text based on the word dictionary and the grammar rules for generation, and outputs the text.

Here, the text analysis unit 21, the language conversion unit 22, and the text generation unit 23 perform processing using the prosody data output from the speech recognition unit 1 as necessary. When the text analysis unit 21, the language conversion unit 22, and the text generation unit 23 perform processing using prosody data, a processing result including prosody data is output as necessary. If the speech recognition unit 1 does not always output prosody data, the text analysis unit 21 and the language conversion unit 2
2, and the text generation unit 23 outputs a request signal requesting the prosody information to the speech recognition unit 1 when the prosody information is required.

Further, the text analysis unit 21 can hold information of the dictionary database 24 and the grammar database for analysis 25 referred to during the processing. Similarly, the language conversion unit 22 and the text generation unit 23 also refer to the information of the language conversion database 26 referred to during the processing,
Dictionary database 27 and grammar database 2 for generation
8 information can be stored. In this case, in the text analysis unit 21, the language conversion unit 22, and the text generation unit 23, a request signal inquiring which information is referred to for processing is sent from a block (for example, the speech synthesis unit 3) that performs processing later. When sent,
Information (eg, prosody data or the like) requested by the request signal can be returned to the block.

Next, FIG. 6 shows language conversion data stored in the language conversion database 26 of FIG.

In this embodiment, the correspondence for translating the source language into the target language is described as the language conversion data together with the prosody information of at least one of the source language and the target language. Language conversion data is registered in the conversion table.

That is, FIG. 6 shows a conversion table when one of Japanese or English is used as a source language and the other is used as a target language.

In the conversion table of FIG. 6 (A), as the text in the source language, a bilingual translation of the text “English teacher” and “English teacher” in the target language expressed as “English teacher” is used. It is described using a symbol * indicating the highlighted part as prosody information. That is, in FIG. 6A, English "English * teac
Japanese "English teacher" is associated as a translation of her * ", and Japanese" English teacher "is associated as a translation of English" * English * teacher ".

In this case, the language conversion unit 22 refers to the conversion table to make the utterance “English * teache
"r *" can be accurately translated into Japanese "English teacher" using the speech recognition result and prosody data. The language conversion unit 22 outputs the utterance in English "*
English * teacher "can be accurately translated into Japanese" English teacher "using the speech recognition result and prosody data.

In the conversion table shown in FIG.
Japanese "NP is VP" as a translation of Japanese "NP is VP" as a translation of "VP"
Are associated with each other. Here, NP is a noun phrase (Noun Ph
rase), and VP indicates a verb phrase (VerbPhrase). In this case, the language conversion unit 22 outputs the utterance "NPV
P ", using the speech recognition result and prosody data,
Japanese "NP is VP" can be translated accurately,
The English utterance "* NP * VP" can be accurately translated into Japanese "NP VP".

In the conversion table of FIG. 6C, Japanese translations of "NP is VP" and "NP is VP"
English "NP VP" is associated, and Japanese "* NP * is V
The English translation "* NP * VP" is associated with the translation of "P". In this case, in the language conversion unit 22, Japanese "NP is VP"
Or "NP VP" is translated as English "NP VP".
The speech synthesizer 3 generates a synthesized sound corresponding to the English “NP VP”, that is, a synthesized sound without accent. The language conversion unit 22 outputs English “* NP * VP” as a translation of Japanese “* NP * is VP”.
A synthetic sound corresponding to the English language "* NP * VP", that is, the noun phrase NP
A synthesized sound in which accents are added to is generated. Therefore,
As a translation result, a synthesized sound without a sense of incongruity is output.

In the conversion table of FIG. 6D, the English “NP1
Also as a translation of V * NP2 *. ", the Japanese" NP1 is also NP2
V. "Is associated with English" * NP1 * also V NP2. "
The Japanese translation "NP1 is also NP2 V." Here, V represents a verb. Also in this case, the language conversion unit 22 outputs the utterance "NP VP""NP1 also V * NP
2 *. "Can be accurately translated into Japanese" NP1 is also NP2 V. "Also, English utterance" * NP1 * also V NP2. "
Can be accurately translated into Japanese "NP1 also NP2 V."

In the embodiment shown in FIG. 6, an asterisk is used to indicate emphasis (representing a phrase that is strongly pronounced). However, to express emphasis or other prosody information, for example, HTML is used. (Hyper Text Markup Languag
It is also possible to use tags adopted in e) and the like.
However, even if attention is paid only to accents in prosody information, for example, there are high and low types and high and low types depending on the language, and it is troublesome to describe prosody information using tags that distinguish such types. . Therefore, when creating a conversion table, using a symbol such as an asterisk that does not distinguish such a type, a description of a simplified conversion table is made, so to speak, using a conversion tool or the like,
The simplified conversion table can be converted into a conversion table described using a tag.

Next, FIG. 7 shows an example of the configuration of the speech synthesizer 3 in FIG.

The text analysis unit 31 receives a text as a translation result output from the machine translation unit 2 as an object of the speech synthesis processing. The text is analyzed with reference to the usage grammar database 35.

That is, the dictionary database 34 stores a word dictionary in which part-of-speech information of each word (phrase) and information such as reading and accent are described, and the analysis grammar database 35 stores a dictionary. For words described in the word dictionary of the database 34, grammatical rules for analysis such as restrictions on word chains are stored. Then, the text analysis unit 31 performs an analysis such as morphological analysis or syntax analysis of the text input thereto based on the word dictionary and the grammatical rules for analysis, and performs a rule speech synthesis performed by the rule synthesis unit 32 in the subsequent stage. Extract necessary information. Here, the information necessary for the rule speech synthesis includes, for example, information for controlling the position of a pause, accent and intonation, other prosody information, and phoneme information such as pronunciation of each word.

The information obtained by the text analysis unit 31 is supplied to the rule synthesis unit 32. The rule synthesis unit 32 uses the phoneme segment database 36 to generate synthesized speech corresponding to the text input to the text analysis unit 31. Is generated.

That is, the speech segment database 36 stores speech segment data in the form of, for example, CV (Consonant, Vowel), VCV, CVC, and the like.
Based on the information from the text analysis unit 31, necessary speech segment data is connected, and a pause, an accent, an intonation, and the like are appropriately added, so that the synthesized speech corresponding to the text input to the text analysis unit 31 is obtained. To generate audio data.

This audio data is supplied to the DA converter 33, where it is converted into an audio signal as an analog signal. This audio signal is supplied to a speaker (not shown),
As a result, a synthesized sound corresponding to the text input to the text analysis unit 31 is output.

Here, the text analyzing unit 31, the rule synthesizing unit 32, and the DA converting unit 33 include the prosody data output by the speech recognizing unit 1 and the translation result output by the machine translating unit 2 as necessary. The processing is performed using the prosody data. Note that the voice recognition unit 1
If the prosody data is not always output, the text analyzing unit 31, the rule synthesizing unit 32, and the DA converting unit 33 perform
A request signal for requesting the request is output to the voice recognition unit 1. Similarly, if the machine translation unit 2 does not always output a processing result including prosody data, the text analysis unit 31, the rule word end unit 32, and the DA conversion unit 33 output the prosody information. When necessary, a request signal requesting the request is output to the machine translation unit 2.

In this embodiment, the voice synthesizing unit 3
Can process text described using a tag representing prosody information (hereinafter, appropriately referred to as a prosody tag). For example, if your emphasis is expressed as a start tag <strong> and an end tag </ strong>
In the speech synthesis unit 3, the part "emphasized" is used as a synthesized sound corresponding to a text such as "partially emphasize <strong>speak" described using a prosody tag. Is output.

In the DA converter 33, the prosody data is used to adjust, for example, the utterance speed and volume of the synthesized sound.

Next, referring to the flowchart of FIG.
The operation of the speech translation system in FIG. 2 will be further described.

When a speech in the source language is input to the speech translation system of FIG. 2, the speech is recognized by the speech recognition unit 1.

That is, in the voice recognition unit 1 (FIG. 4), the voice in the original language is input to the microphone 11 and further converted into digital voice data via the AD conversion unit 12. This audio data is supplied to the audio data buffer 14A and is temporarily stored, and the characteristic amount extraction unit 13
Supplied to In the feature value extraction unit 13, in step S1, feature parameters are extracted from the audio data from the A / D conversion unit 12, supplied to the feature value buffer 14B and temporarily stored, and supplied to the matching unit 15 in step S2. Proceed to.

In step S2, the matching unit 15 performs speech recognition of the source language speech (input speech) based on the feature parameters from the feature extraction unit 13, and generates a text or the like in the source language representing the speech recognition result. Is done.

The matching unit 15 can output a plurality of voice recognition result candidates without specifying one voice recognition result. In addition, the matching unit 15 can output a speech recognition result in the form of a word graph having ambiguity, for example. Here, for example, the voice in Japanese "Near this,
Examples of word graphs for "Is there a good restaurant?" Are shown in FIGS. 9 and 10. In the embodiment of FIGS. 9 and 10, the start time and the end time based on the time at which the utterance is started, a candidate for a speech recognition result (word candidate) between the start time and the end time, and The log likelihoods of the word candidates are arranged in chronological order. FIG. 10 is a view following FIG.

Returning to FIG. 8, in step S2, further,
In the prosody information extracting section 19, prosody information of the input audio is extracted based on the audio data stored in the audio data buffer 14 and the feature amount stored in the feature amount buffer 14B, and the prosody data representing the prosody information is extracted. Are output to the matching unit 15.

Here, in the present embodiment, for example, prosodic information, information that reflects the personality of the user (the user who uttered the input voice), or information that has been abstracted is extracted as prosody information. It has become. Primitive prosody information, for example,
There are pitch, power, utterance speed, utterance time, and pause length of the input voice. In addition, prosody information reflecting personality of a user or abstracted prosody information includes strong / slow accent / intonation, high / low accent / intonation, male or female voice, and any emotional tone. Something (eg, angry tone, laughing tone, etc.), non-verbal sound information (eg, tongue and sneeze information), declarative text, or question text There are, for example, word break positions.

In the prosody information extracting section 19, the primitive prosody information is obtained by performing an operation using audio data and its characteristic parameters, and the prosody information reflecting the personality of the user or the abstracted prosody information is obtained. Is obtained using one or more pieces of primitive prosody information.

That is, for example, the position of the break of a word is determined using the length of the pause at that position or the change in the height intonation. Specifically, for example, the Japanese word "All lockers are paid" can mean "I will provide lockers to everyone" if you look at the text alone, and there is something called "All lockers, Will be provided. " In this case, if the Japanese word "all lockers are supplied" means, for example, "pay lockers to all", the Japanese language "all" and "lockers" will be used. Is a delimiter position of a phrase.

On the other hand, when the utterance “all lockers are supplied” means “provide lockers to all”, “all” and “lockers” in the utterance “all lockers are supplied” are used. , A pause is generally inserted, and the intonation between them changes from low to high.

Therefore, in this case, by using the primordial prosody information such as the presence / absence of a pause between the “all” and the “rocker” in the utterance “all lockers are supplied” and the intonation, the position between them can be used to separate words and phrases. The position can be determined.

[0109] By using the information of the word delimiter position of the speech recognition result in the machine translator 2, it is possible to prevent mistakes in the parsing of the speech recognition result when parsing the speech recognition result. Can be.

Here, FIG. 11 shows prosody data output by the prosody information extracting unit 19.

In FIG. 11A, high / low accents, strong / slow accents, and phrase delimiters as prosody information are described together with respective start times and end times, and the degree of high / low accents and strong / slow accents is also described. It has been described. The degree of high / low accents and strong / slow accents is represented by a plus number when it is high or strong and a minus number when it is low or weak, based on 0.

In FIG. 11B, only the start time and the end time of the high accent as prosody information are described. If the machine translation unit 2 and the speech synthesis unit 3 that perform processing in the subsequent stage do not need all of the prosody information as described above but need only the presence or absence of a high accent higher than a certain height, the prosody information extraction is performed. It is useless to extract all pieces of prosody information in the unit 19, and only necessary information need be extracted. Further, if the required prosody information is only the high accent as described above and only the presence or absence of a high accent of a certain height or higher, as shown in FIG. The type and degree are not required. Therefore, in such a case, as shown in FIG. 11B, prosody data describing only the start time and the end time in which prosody information of a high accent equal to or higher than a certain height exists can be used. It is.

Note that the prosody data is included in the speech recognition result in the matching unit 15 as described below, but is converted into a prosody tag at that time. FIG. 11C shows an example of a prosody tag.

Returning again to FIG. 8, the matching unit 15
Upon receiving the prosody data from the prosody information extracting unit 19, in step S 3, a recognition result with a prosody tag including the prosody data in the speech recognition result is generated and output to the machine translation unit 2.

In the machine translation unit 2 (FIG. 5), step S4
, The machine translation into the target language is performed using the recognition result with the prosody tag from the speech recognition unit 1.

That is, in the machine translation unit 2, the text analysis unit 21 performs a text analysis of the speech recognition result of the utterance in the original language using the recognition result with the prosody tag from the speech recognition unit 1, and the analysis result Is the language conversion unit 2
2 is supplied. Here, by referring to the prosody tag in the text analysis unit 21, it is possible to prevent the above-described analysis in which an erroneous translation is performed from being performed.

The language conversion unit 22 translates the recognition result with the prosody tag of the source language by referring to the language conversion data, and converts it into the translation result with the prosody tag of the target language. The translation result with the prosody tag is supplied to the text generation unit 23, where it is processed and then output to the speech synthesis unit 3 and the display unit 4.

In the voice synthesizing section 3 (FIG. 7), step S5
In, the translation result with the prosody tag from the machine translation unit 2 is analyzed. That is, the text analysis unit 31 of the speech synthesis unit 3 analyzes the translation result with the prosody tag from the machine translation unit 2 with high accuracy by referring to the prosody tag, and obtains phonological information and speech information required for the rule-based speech synthesis. , And a symbol string composed of accent marks or the like representing accents for performing rule-based speech synthesis. That is, the character part of the translation result with the prosody tag is converted into phoneme information,
Also, for example, a prosody tag or the like indicating that it is emphasized and pronounced is converted into a predetermined symbol string.

The symbol string output by the text analysis unit 31 is supplied to the rule synthesizing unit 32. In step S6, the rule synthesizing unit 32 performs rule-based speech synthesis based on the symbol string. Here, the symbol string output by the text analysis unit 31 includes a prosody tag that was not converted by the text analysis unit 31 and that was included in the translation result with the prosody tag.
Processing is performed using the prosody tag included in the symbol string as needed. As a result, the rule speech synthesizer 32 generates data such as a synthesized speech at substantially the same utterance speed of the input speech, a gender of the user who uttered the input speech, or a synthesized tone reflecting the tone. This data is converted into an analog signal in the DA converter 33,
As a result, a synthesized sound corresponding to the translation in the target language is output from the speaker 5. In the display unit 4,
The translation in the target language supplied from the machine translation unit 2 is displayed, and the process ends.

According to the above-described processing, it is possible to perform high-accuracy translation and to output a translated sentence as a synthesized sound without a sense of incongruity.

That is, for example, the utterance in the source language is “I a
lso like * her *. ", the matching unit 15 of the voice recognition unit 1 obtains, for example, a word graph as shown in FIG. For example, assume that prosody data as shown in Fig. 12B is obtained at step 19. In the embodiment of Fig. 12B, Fig. 11
As in the case described in (B), prosody data in which only the start time and the end time of the prosody information where a strong accent exists is used.

In this case, the matching unit 15 generates a recognition result with a prosody tag as shown in FIG. 12C from the word graph and the prosody data. Here, the recognition result with prosody tag “I also like <stress> her </ stress>.” Shown in FIG. 12C is a start tag <stress> and an end tag </ stress of a prosody tag representing a strong accent. This indicates that the word "her" between and> has a strong accent.

The recognition result with the prosody tag as shown in FIG. 12C is supplied to the machine translation unit 2, and after the text analysis unit 21 analyzes the result, the language conversion unit 2
2 is supplied. In the language conversion unit 22, matching between the recognition result with the prosody tag and each language conversion data in the language conversion database 26 is performed.
Language conversion data that best matches the recognition result with the prosody tag is detected.

That is, it is assumed that the conversion table shown in FIG. 6 is stored in the language conversion database 26 referred to by the language conversion unit 22, and the text analysis unit 21
Recognition result with prosody tag by "I also like <stre
ss> her </ stress>. "
ress> NP2 </ stress>. "

In this case, the recognition result with prosody tag “I
also like <stress> her </ stress>. "
so V <stress> NP2 </ stress>. "is" NP1 also V * NP2 *. "described in the first line of the conversion table in FIG. Language converter 22
Then, the language conversion data is detected. That is, the first word "NP1" of the analysis result and the first word "NP" of the language conversion data
1 ", the second word" also "of the analysis result and the second word" also "of the language conversion data, the third word" V "of the analysis result and the third word" V "of the language conversion data, and Analysis result 4
The fourth (last) word "NP2" and the fourth (last) word "NP2" in the language conversion data match, and the prosody tag <stress surrounding the fourth word "NP2" in the analysis result
> And </ stress> and the fourth word "N" in the language conversion data
The asterisks surrounding P "all indicate emphasis, so they also match. Therefore, in the language conversion unit 22, FIG.
"NP1a" described in the first line of the conversion table of (D)
lso V * NP2 *. "is the recognition result with prosody tag" I also
like <stress> her </ stress>. "
<stress> NP2 </ stress>. ".

Then, the language conversion unit 22 converts the detected language conversion data "NP1 also V * NP2 *."
2 also V. "Prosody-tagged speech recognition results" I also
like <stress> her </ stress>. "to the text generator 23.

When the prosody tag of the recognition result with the prosody tag is ignored, that is, when the speech recognition result without the prosody tag is analyzed and the language conversion data that best matches the analysis result is detected, the diagram shown in FIG. 6 (D) "N
"P1 also V * NP2 *." And "* NP1 * also V NP2." Are both detected, and it is difficult to determine the superiority. Therefore, it is difficult to continue the subsequent processing. .

As a method of matching the recognition result with the prosody tag and the language conversion data, for example, a search method using a logical expression, a search method using a vector space method, and the like, which are used in general sentence search, are employed. can do. When a search method using a logical expression is adopted, a logical expression (for example, a word that becomes a keyword is AND, OR, N
OT or the like is created, and by comparing the logical expression with the language conversion data, the language conversion data that best matches the recognition result with the prosody tag is retrieved. When a search method based on the vector space method is adopted, a query vector (Query Vector) is created from words constituting the recognition result with the prosody tag, and a vector is created based on the language conversion data. The language conversion data that minimizes the angle formed by the two vectors is searched for as the language conversion data that best matches the recognition result with the prosody tag.

The text generator 23 converts the language conversion data "N
P1 also V * NP2 *. "And its translation" NP1 is also NP2 V. "and the speech recognition result with prosody tag" I also like <stre
ss> her </ stress>. ", the" I "corresponding to NP1," like "corresponding to V, and" her "corresponding to NP2 in the language conversion data are converted into the target language Japanese. By translation, this gives “I” for NP1, “I like” for V, and “She” for NP2. Then, the text generation unit 23 applies those Japanese translations to the bilingual translation “NP1 is also NP2 V.”
As shown in 2 (D), "I like her."

This translation result "I like her too."
Is supplied to the speech synthesis unit 3, whereby “I also li
As a translation result for "ke * her *.", a synthetic sound "I like her" is output.

In the speech translation system shown in FIG. 2, for example, as described in the above-mentioned Japanese Patent Application Laid-Open No. 6-332494, a phrase in the target language corresponding to an accented phrase in the input speech in the source language is used. It is also possible to output with accents.

That is, for example, the above-mentioned input voice “I also li
For "ke * her *.", it is possible to detect that "her" has an accent from its prosody-tagged recognition result "I also like <stress> her </ stress>." During translation, it can also be detected that the Japanese translation corresponding to the accented "her" is "her". Therefore, by adding a prosody tag indicating emphasis to “her” of the translation result “I also like her.” And outputting the translation result with the prosody tag to the speech synthesizer 3, the input speech in the original language can be obtained. Thus, it is possible to output a synthesized sound in which an accent is added to a word in the target language corresponding to the accented word.

However, since accents in English are strong and weak accents, accents in Japanese are high and low accents, the prosody tag added to the translation result is the recognition result with prosody tag "I also
Like <stress> her </ stress>. ", it is preferable to use a prosody tag indicating a high / low accent instead of a prosody tag indicating a high / low accent. As shown in FIG. 12, when expressed by <hi-picth> and </ hi-picth>, as shown in FIG. 12 (E), the translation result with the prosody tag is “I <hi-picth> her </ I also like hi-picth>. "
Becomes

As described above, in the machine translation unit 2, the translation information for translating the source language into the target language is based on the conversion table in which at least one of the source language and the target language is described. The translation is performed with high accuracy.
It is possible to output a translated sentence with a synthetic sound without a sense of incongruity (it is possible to output a synthetic sound suitable as a translation result).

That is, for example, as shown in FIG. 13, when an English user utters “I also like * her *.”, As shown in FIG. 14, “I also like her. Is output. FIG.
3 (similarly in FIG. 15 described below),
Words to be emphasized are underlined.

For example, as shown in FIG. 15, when an English user utters “* Henry * has arrived.”, The correct translation result “English has arrived.” In consideration of the accent of the utterance, the synthesized sound of the translation using the appropriate particle "ga") is output with the correct position accented.

Further, for example, as shown in FIG. 16, when the English user utters "Henry has arrived.", The correct translation result of "Henry has arrived." Given that there is no translation, the synthesized sound of the translation using the appropriate particle "ha")
Outputs with correct accents.

For example, while “English * teacher *” is correctly translated to “English teacher”, “* Engl
Correct translation of "ish * teacher" to "English teacher" requires the dictionary database 1 in the speech recognition unit 1 (FIG. 4).
For 7th grade, a dictionary describing "English * teacher *" and "* English * teacher" as different words (phrases) is registered, and "English * teacher *" and "* English * teacher" In addition to outputting the speech recognition result separately, the machine translation unit 2 also outputs "English * teacher *" and "* English
* Teacher's voice recognition results can be handled separately.

However, although they are the same word string,
In order to create a dictionary for performing speech recognition by distinguishing voices with different accented words, for example, a word sequence in which accented words move by compounding and the translation result is It is necessary to investigate what is different from the case where a word has not moved, and write it in the dictionary. Therefore, when creating such a dictionary, the creator who manufactures the speech recognition unit 1 is required to have linguistic knowledge other than knowledge for speech recognition processing. It is considered difficult.

In the above case, the prosody information extracting section 19 of the voice recognition section 1 (FIG. 4) always extracts prosody information and outputs prosody data. Not all of them are used in the machine translation unit 2 and the speech synthesis unit 3, and only some of them are sometimes used (from the input speech in the source language described in the above-mentioned Japanese Patent Application Laid-Open No. 6-332494). In a translation device that extracts words with accents and adds accents to the words in the target language corresponding to the words, the accent is also extracted over the entire voice section, but is reflected in the synthesized speech of the translation result. It is part of the extracted accent).

Specifically, for example, as shown in FIGS. 15 and 16, the utterance “* Henry * h
When translating between "as arrived." and "Henry has arrive.", the only thing that affects the translation result among the prosody information of each utterance is whether "Henry" has an accent. It is.

Therefore, always extracting prosody information over the entire voice section means extracting prosody information that is not used by the machine translation unit 2 or the speech synthesis unit 3 that performs processing after the speech recognition unit 1. And the processing efficiency is not good.

Therefore, the prosody information extraction unit 19 of the speech recognition unit 1 (FIG. 4) needs a necessary speech segment only when requested by the machine translation unit 2 or the speech synthesis unit 3 that performs processing in the subsequent stage. Only necessary prosody information can be extracted for only the target. In this case, processing efficiency can be improved (the amount of calculation can be reduced).

From the same point of view, the machine translation unit 2 is also made to perform predetermined processing only when requested by the speech synthesis unit 3 which performs processing in the subsequent stage. It is possible to improve processing efficiency.

In this case, the speech recognition unit 1, the machine translation unit 2,
The voice synthesizing unit 3 exchanges data with each other as shown in FIG. In FIG. 17, the vertical direction indicates the passage of time.

In the case shown in FIG. 17, when a voice is input, the voice recognition section 1 performs voice recognition,
The speech recognition result D ₁ is output to the machine translation unit 2.
In Prosody extraction unit 19 of the speech recognition unit 1 at this time, extraction of Prosody is not performed, therefore, the speech recognition unit 1 outputs the voice recognition results D ₁ contains no prosodic data. Here, the prosody information extracting unit 19 does not extract any prosody information, but always extracts only prosody information frequently used in processing in the machine translation unit 2 and the speech synthesis unit 3. , it is possible to be included in the speech recognition result D _1.

[0147] Machine translation unit 2 receives the speech recognition result D ₁ from the speech recognition unit 1 performs processing using the speech recognition result D _1, during the process, when it becomes necessary Prosody Interrupts the process and outputs to the speech recognition unit 1 a request signal D ₂ requesting necessary process information.

Here, the prosody information requested by the request signal D ₂ includes, for example, whether or not a voice corresponding to a certain word has a strong or weak accent (to what degree, Accent), whether the gender of the user (speaker) is male or female, whether there is a speech break between a word and the word immediately following it, and the intonation at the end of the sentence is up or down There is information such as which of the conditions is.

The prosody information extraction unit 19 of the speech recognition unit 1
From the machine translation unit 2, upon receiving the request signal D _2, the prosody information requested by the request signal D _2, determined by referring to the audio data buffer 14A and the feature quantity buffer 14B, the request signal D ₂ in response D ₃ against, and outputs to the machine translation unit 2.

When receiving the response D ₃ , the machine translator 2
Continue the process by using the prosody information included in the response D _3, hereinafter, during processing, if it becomes necessary Prosody suspends the processing, a request signal requesting the Prosody, It outputs to the voice recognition unit 1, obtains a response as necessary prosody information from the voice recognition unit 1, and restarts the process.

When the translation of the speech recognition result D ₁ from the speech recognition unit 1 is completed, the machine translation unit 2 outputs the translation result D ₄ to the speech synthesis unit 3.

Upon receiving the translation result D ₄ from the machine translation unit 2, the speech synthesis unit 2 performs processing using the translation result D _4, and during the processing, prosody information and other information are required. In this case, the processing is interrupted and the request signal D ₅ requesting necessary prosody information and other information is interrupted.
The output to the machine translation unit 2, or outputs the same request signal D ₇ to the speech recognition unit 1.

Here, as information requested by the machine translation unit 2 by the request signal D ₅ , for example, pronunciation of the same transcription word (for example, “read” is pronounced differently between the present tense and the past tense) However, its pronunciation), the accent of the same spelling different accent word (for example, "increase" has a different accent depending on the part of speech, but the accent position), and the intonation of a word string that may be compounded (for example, There is information such as whether the intonation at the end of the sentence is in ascending or descending tone. Please note that this information
It is also possible to always include the same text as the prosody tag in the text as the translation result output by the machine translator 2.

Information requested by the voice recognition unit 1 by the request signal D ₇ includes the speech speed and pitch of the input voice, the user's gender, emotion, tone, and the intonation of the end of the sentence, either in ascending or descending tone. There is information such as whether there is.

Upon receiving the request signal D ₅ from the speech synthesizing unit 3, the machine translating unit 2 converts the information requested by the request signal D ₅ into the text analyzing unit 21 and the language converting unit as described above. 22, text generator 23
By referring to the information that was retained during processing,
In response D ₆ to the request signal D _5, and outputs the speech synthesizing unit 3.

Further, upon receiving the request signal D ₇ from the voice synthesizing unit 3, the voice recognizing unit 1 transmits, for example, prosody information requested by the request signal D ₂ to the voice data buffer 14 A and the feature amount buffer 14 B. determined by reference to the, as a response D ₈ to the request signal D _7, and outputs the speech synthesizing unit 3.

Upon receiving the responses D ₆ and D ₈ , the speech synthesizing unit 3 continues the process using the information included in the responses D ₆ and D _8. Hereinafter, during the processing, the prosody information and other necessary a case where the information is required, interrupts the process, a request signal D ₅ and D ₇ to request the necessary information, and outputs to the machine translation unit 2 and the speech recognition unit 1, the machine translation unit 2 And a response as necessary information is obtained from the voice recognition unit 1 and the process is repeated.

It should be noted that the request signal is not supplied directly from the speech synthesis section 3 to the speech recognition section 1 but is supplied indirectly via the machine translation section 2. Also, instead of directly supplying a response from the speech recognition unit 1 to the speech synthesis unit 3,
It is possible to supply a response indirectly via the machine translator 2.

That is, for example, the input voice "I also like * he
r *. ", the Japanese translation of which is" I also like her. "
If you output the synthesized sound of the word "her" that is emphasized in the input voice, such as "I also like * her *."
When emphasizing “She” corresponding to, the voice synthesis unit 3
Receives a translation from "Machine Translator 2" saying "I also like her" and then sends a request signal requesting a check to see if the input speech corresponding to "her" in it has an accent. , To the machine translation unit 2. The machine translation unit 2 relates to the request by the request signal,
Since it recognizes that the word in the input speech corresponding to "her" is "her", the request signal requesting whether the input speech corresponding to the "her" has an accent is recognized by speech recognition. Output to section 1. Voice recognition unit 1
Then, "her" in the input voice "I also like * her *."
Is checked for an accent, and in this case, because "her" has an accent,
The request signal is supplied to the machine translator 2 which has output the request signal. When the machine translation unit 2 receives a response indicating that "her" has an accent from the voice recognition unit 1, the machine translation unit 2 transmits a response indicating that the input voice corresponding to "her" has an accent to the voice that has output the request signal. Output to the combining unit 3. In this way, the speech synthesis unit 3 outputs “her” from the machine translation unit 2.
Receives a response indicating that the input speech corresponding to has an accent, generates a synthetic sound that emphasizes "her" in the translated sentence "I also like her." Output.

Next, referring to the flowchart of FIG. 18, machine translation unit 2 and speech synthesis unit 3 output a request signal, and speech recognition unit 1 and machine translation unit 2 output a response to the request signal. The operation of the voice recognition unit 1 in such a case will be described.

In the voice recognition unit 1, in step S11, a waiting time is set until an event occurs, such as whether a voice is input or a request signal is transmitted from the machine translation unit 2 or the voice synthesis unit 3. When an event occurs, the process proceeds to step S12, and it is determined what event has occurred. In step S12,
If it is determined that an event that a voice has been input has occurred, the process proceeds to step S13, the input voice is recognized, and the process proceeds to step S14. In step S14, the speech recognition result in step S13 is output to the machine translator 2, and the process returns to step S11, and thereafter, the same processing is repeated. In step S14, the speech recognition result may be output to the machine translation unit 2 and, if necessary, supplied to the display unit 5 for display.

On the other hand, if it is determined in step S12 that an event that a request signal has been transmitted has occurred, the process proceeds to step S15, and prosody information is extracted in response to a request by the request signal.
That is, in step S15, the prosody information extraction unit 19
In, for example, extraction of accents, calculation of speech rate, identification of whether the user is male or female,
Non-verbal sound information is extracted. Then, the process proceeds to step S16, in which prosody data representing the extracted prosody information is provided as a response to the request signal.
The request signal is supplied to the block that has transmitted the request signal (in this case, either the machine translation unit 2 or the speech synthesis unit 3), the process returns to step S11, and the same process is repeated.

Next, referring to the flowchart of FIG. 19, the machine translation unit 2 and the speech synthesis unit 3 output a request signal, and the speech recognition unit 1 and the machine translation unit 2 output a response to the request signal. In this case, the operation of the machine translation unit 2 will be described.

At step S21, the machine translator 2 waits until an event such as whether a speech recognition result has been transmitted from the speech recognizer 1 or a request signal has been transmitted from the speech synthesizer 3 has occurred. When an event occurs, the process proceeds to step S22,
It is determined what event has occurred. If it is determined in step S22 that the event that the speech recognition result has been transmitted has occurred, the process proceeds to step S23, where the speech recognition result is analyzed by the text analysis unit 21, and the analysis result is transmitted to the language conversion unit 22. Supplied. In step S24, the language conversion unit 22 converts the speech recognition result in the source language into language information in the target language based on the analysis result output by the text analysis unit 21 and outputs the result to the text generation unit 23 in step S24. In step S25, the text generation unit 23 translates the speech recognition result in the source language into a text in the target language based on the language information from the language conversion unit 22, and proceeds to step S26. Output to the unit 3. Then, the process returns to step S21, and thereafter, the same processing is repeated.

In step S26, the translation result can be output to the speech synthesis unit 3 and, if necessary, supplied to the display unit 5 for display. Further, the text analysis unit 21 and the language conversion unit 2
As described later, if the information (here, prosody information) required by the voice recognition unit 1 becomes necessary during the processing, the text generation unit 23 interrupts the processing and outputs the information. A request signal is transmitted to the voice recognition unit 1. Then, in response to the request signal, the process is restarted after information is transmitted from the voice recognition unit 1.

On the other hand, if it is determined in step S22 that an event that a request signal has been transmitted from the voice synthesizing unit 3 has occurred, the process proceeds to step S27, and information requested by the request signal is obtained. That is, as described above, the machine translation unit 2
It holds information of the dictionary database 24, the grammar database for analysis 25, the language conversion database 26, the dictionary database 27, and the grammar database 28 for generation referred to during the processing, and refers to the held information. For example, information required by the request signal, such as the position of an accent in an input voice or its translation result, pronunciation of a word, and nonlinguistic information, is obtained.
Then, the process proceeds to step S28, where the obtained information is supplied as a response to the request signal to the voice synthesizing unit 3 that has transmitted the request signal.
Returning to 1, the same processing is repeated thereafter.

Next, with reference to FIG. 20, the details of each of the processes in steps S23 to S25 in FIG. 19 will be described.

First, the details of the processing in step S23 performed by the text analysis unit 21 of the machine translation unit 2 will be described with reference to the flowchart in FIG.

In step S31, the text analysis unit 21 reads the input from the speech recognition unit 1, that is, reads all or a part of the speech recognition result, and proceeds to step S32.
It is determined whether some kind of prosody information is needed to process the speech recognition result. In step S32, in particular, when it is determined that prosody information is not necessary, steps S33 and S34 are skipped, and the process proceeds to step S35, where the speech recognition result is analyzed.

If it is determined in step S32 that some kind of prosody information is necessary, the process proceeds to step S33, where the text analysis unit 21 requests a request signal (for which voice section, which A signal including whether such kind of prosody information is required) is output to the voice recognition unit 1. Then, after waiting for the necessary prosody information corresponding to the request signal to be transmitted from the voice recognition unit 1,
Proceed to step S34. In step S34, necessary process information as a response to the request signal transmitted from the voice recognition unit 1 is received, and the process proceeds to step S35, where the voice recognition result is analyzed using the process information.

After the processing in step S35, step S3
Then, it is determined whether or not all the analysis of the speech recognition result output by the speech recognition unit 1 has been completed (step 6). Step S
In 36, if it is determined that all the analysis of the voice recognition result has not been completed, the process returns to step S31, the voice recognition result that has not been analyzed is read, and the same processing is repeated thereafter.

On the other hand, if it is determined in step S36 that all the analysis of the speech recognition result has been completed, the process returns.

Next, the details of the processing in step S24 performed by the language conversion unit 22 of the machine translation unit 2 will be described with reference to the flowchart in FIG.

In step S41, the language conversion unit 22 reads all or a part of the analysis result of the speech recognition result from the text analysis unit 21, and proceeds to step S42.
It is determined whether any prosody information is needed to process the analysis result. In step S42,
In particular, if it is determined that prosody information is not required,
Skipping steps S43 and S44, the process proceeds to step S45, where the analysis result from the text analysis unit 21 is subjected to language conversion processing.

If it is determined in step S42 that some kind of prosody information is necessary, the process proceeds to steps S43 and S44, and the language conversion unit 22 performs the processing in steps S33 and S34 of FIG. A similar process is performed, whereby necessary prosody information is obtained from the voice recognition unit 1. Then, the process proceeds to step S45, and the analysis result read in step S41 is subjected to language conversion processing using the prosody information.

After the processing in step S45, step S4
Then, it is determined whether or not all the language conversions of the analysis result output by the text analysis unit 21 have been completed (step 6). In step S46, all language conversions of the analysis result are
If it is determined that the processing has not been completed, step S41
Then, the analysis result that has not been subjected to language conversion is read, and the same processing is repeated thereafter.

On the other hand, if it is determined in step S46 that all the language conversions of the analysis result have been completed, the process returns.

Next, the details of the processing in step S25 performed by the text generation unit 23 of the machine translation unit 2 will be described with reference to the flowchart in FIG.

In step S51, the text generation unit 23 reads all or a part of the language conversion result from the language conversion unit 22, and proceeds to step S52, where some sort of prosody information is required to process the language conversion result. Is determined. In step S52,
If it is determined that prosody information is not necessary, steps S53 and S54 are skipped and step S55 is performed.
The text of the target language is generated for the language conversion result read in step S51.

If it is determined in step S52 that some kind of prosody information is necessary, the flow advances to steps S53 and S54, and the text generation unit 23 executes the processing in steps S33 and S34 of FIG. A similar process is performed, whereby necessary prosody information is obtained from the voice recognition unit 1. Then, the process proceeds to step S55, and the target language text is generated using the prosody information and targeting the language conversion result read in step S51.

After the processing in step S55, step S5
Proceeding to 6, it is determined whether or not text generation has been completed for all of the language conversion results output by the language conversion unit 22. If it is determined in step S56 that the generation of the text for all of the language conversion results has not been completed yet, the process returns to step S51, where the language conversion results for which the text has not been generated are read.
A similar process is repeated.

On the other hand, if it is determined in step S56 that the generation of texts for all of the language conversion results has been completed, the process returns.

Next, referring to the flowchart of FIG. 21, machine translation unit 2 and speech synthesis unit 3 output a request signal, and speech recognition unit 1 and machine translation unit 2 output a response to the request signal. The operation of the voice synthesizing unit 3 in the case of performing the operation will be described.

The speech synthesizing unit 3 waits in step S61 until an event of transmission of the translation result from the machine translation unit 2 occurs. When the event occurs, the process proceeds to step S62, where the machine translation unit 2 Is translated in the text analysis unit 31, and the analysis result is supplied to the rule synthesis unit 32. Rule synthesis part 3
2, in step S63, the text analysis unit 31
The rule-based speech synthesis is performed based on the analysis result output by, and digital data of synthesized speech is generated. This digital data is supplied to the DA converter 33, which converts the digital data from the regular voice synthesizer 32 into digital data in step S64, and thereby generates a synthesized sound corresponding to the translation result. Are output from the speaker 5. Then, the process returns to step S61, and thereafter, the same processing is repeated.

Note that the text analyzer 31 and the rule synthesizer 3
If the information required by the speech recognition unit 1 or the machine translation unit 2 becomes necessary during the processing, the DA conversion unit 33 suspends the processing and sends a request signal requesting the information to the speech It is transmitted to the recognition unit 1 and the machine translation unit 2.
Then, in response to the request signal, the speech recognition unit 1
And waiting for information to be sent from the machine translation unit 2,
Resume processing.

Next, with reference to FIG. 22, the details of the processing in steps S62 and S63 in FIG. 21 will be described.

First, the details of the processing in step S62 performed by the text analysis unit 31 of the speech synthesis unit 3 will be described with reference to the flowchart in FIG.

In step S71, the text analysis unit 31 reads the input from the machine translation unit 2, that is, reads all or a part of the translation result, and proceeds to step S72. (And other information) is determined. In step S72, if it is determined that prosody information is not particularly necessary, steps S73 and S74 are skipped, and the process proceeds to step S75, where the translation result is analyzed.

If it is determined in step S72 that some kind of prosody information is necessary, the process proceeds to step S73, where the text analysis unit 31 requests a request signal (for which voice section, which A signal including whether such kind of prosody information is required) is output to the speech recognition unit 1 or the machine translation unit 2. Then, the process waits until the necessary prosody information as a response corresponding to the request signal is transmitted from the voice recognition unit 1 or the machine translation unit 2, and then the process proceeds to step S74.
Proceed to. In step S74, necessary prosody information as a response to the request signal transmitted from the voice recognition unit 1 or the machine translation unit 2 is received, and the process proceeds to step S75, where the translation result is obtained using the prosody information. Is parsed.

After the processing in step S75, step S7
Then, it is determined whether the analysis of all the translation results output by the machine translation unit 2 has been completed or not. Step S76
In, when it is determined that all the analysis of the translation result has not been completed, the process returns to step S71, the translation result that has not been analyzed is read, and the same processing is repeated thereafter.

On the other hand, if it is determined in step S76 that all the translation results have been analyzed, the process returns.

Next, the details of the processing in step S63 performed by the rule synthesizing section 32 of the voice synthesizing section 3 will be described with reference to the flowchart in FIG.

In step S81, the rule synthesizing unit 32 reads all or a part of the analysis result of the translation result from the text analysis unit 31, and proceeds to step S82, where some prosody information is used to process the analysis result. Determine if it is necessary. In step S82, in particular, when it is determined that prosody information is not necessary, steps S83 and S84 are skipped, and step S83 is skipped.
Proceeding to 85, rule speech synthesis is performed according to the analysis result read in step S85.

If it is determined in step S82 that some kind of prosody information (further, other information) is necessary, the process proceeds to steps S83 and S84 in order, and the rule synthesizing unit 32 executes the process shown in FIG. Step S7
3, the same processing as in the case of S74 is performed, whereby necessary prosody information is obtained from the speech recognition unit 1 or the machine translation unit 2. Then, the process proceeds to step S85, in which rule speech synthesis is performed according to the analysis result read in step S81 while using the prosody information.

After the processing in step S85, step S8
Proceeding to 6, it is determined whether or not ruled speech synthesis has been completed for all of the analysis results output by the text analysis unit 31. If it is determined in step S86 that the rule-based speech synthesis has not been completed for all the analysis results, the process returns to step S81, where the analysis result for which the rule-based speech synthesis has not been performed is read. Repeated.

On the other hand, if it is determined in step S86 that the ruled speech synthesis has been completed for all the analysis results, the process returns.

As described above, when the speech recognition unit 1 and the machine translation unit 2 provide prosody information in response to the request signal from the machine translation unit 2 and the speech synthesis unit 3, the request Since only the prosody information required by the signal needs to be extracted, the processing efficiency can be improved.

In the present embodiment, the speech recognition result is always transmitted from the speech recognition unit 1 to the machine translation unit 2. However, the speech recognition unit 1 transmits the speech recognition result to the machine translation unit 2. Only when a request signal requesting a required speech recognition result is transmitted from the machine translation unit 2, the necessary speech recognition result can be transmitted.

That is, for example, in the machine translation section 2, the language conversion section 22 performs matching between the speech recognition result and the language conversion data, thereby detecting the language conversion data that best matches the speech recognition result. In this matching, only some words included in the utterance may be required. Specifically, for example, when the language conversion data is classified for each predicate (such as a verb) (for example, “Please give XX” or “Eat XX”)
), The first piece of information needed for matching is
A predicate in the utterance and a phrase related to the predicate. For example, the language conversion data is a particle (for example, “from OO to ×
× ”or“ XX to XX ”), the first information needed for matching is the particles in the utterance and the words placed before and after it. Sometimes only some of the words included in the utterance are needed. It should be noted that a method such as the above-described logical expression or the vector space method can be used to match a part of words in the utterance with the language conversion data.

In this case, the speech recognition unit 1 switches to the machine translation unit 2
It is sufficient to initially supply only the speech recognition results of some phrases included in the utterance, and supply the speech recognition results of the entire utterance after the language conversion data to be used is detected.

Therefore, the speech recognition unit 1 and the machine translation unit 2
In the meantime, the same exchange as in the case described with reference to FIG. 17 can be performed on the speech recognition result.

That is, in this case, the speech recognition section 1 recognizes the utterance by speech and obtains the recognition result.
An event indicating this is transmitted to the machine translation unit 2.
When the machine translation unit 2 receives an event indicating that a speech recognition result has been obtained from the speech recognition unit 1, the machine translation unit 2 outputs a speech request signal for requesting a phrase first necessary for processing the speech recognition result. It is transmitted to the recognition unit 1. Voice recognition unit 1
Receives the request signal from the machine translation unit 2 and transmits the speech recognition result of the phrase requested by the request signal to the machine translation unit 2.

[0203] The machine translation unit 2 analyzes the speech recognition result of the phrase transmitted from the speech recognition unit 1 in response to the request signal. Is detected. Then, the machine translation unit 2 uses the language conversion data to obtain a phrase necessary for performing language conversion (translation) of the speech recognition result, and transmits a request signal requesting the phrase to the speech recognition unit 1. I do.

The speech recognition unit 1 extracts the phrase requested by the request signal transmitted from the machine translation unit 2 from the speech recognition result of the utterance, and transmits the extracted phrase to the machine translation unit 2. Then, the machine translation unit 2 uses the words of the speech recognition result transmitted from the speech recognition unit 1 to generate a translated sentence obtained by converting the speech recognition result into the target language.

As described above, when only the words of the speech recognition result requested by the machine translation unit 2 are transmitted from the speech recognition unit 1 to the machine translation unit 2, the processing efficiency is improved. It is also possible to prevent a reduction in the voice recognition rate.

That is, in the machine translation unit 2, even if the phrase is included in the utterance, the phrase may not be reflected in the translation result, and such a phrase is transmitted from the speech recognition unit 1 to the machine translation unit. As a result, the processing efficiency is improved. In addition, the speech recognition unit 1 may obtain a plurality of candidates for speech recognition results. The machine translation unit 2 may request such a plurality of candidates for speech recognition results as necessary. Then, the voice recognition unit 1 prevents a reduction in the voice recognition rate due to determining and outputting a certain candidate as a final voice recognition result from the plurality of voice recognition result candidates. Becomes possible.

Next, the above-described series of processing can be performed by hardware or software. When a series of processing is performed by software, a program constituting the software is
It is installed in a computer incorporated in a speech translation system as dedicated hardware, a general-purpose computer, or the like.

With reference to FIG. 23, a recording medium on which the program for executing the above-described series of processes is recorded and which is used to install the program in a computer and make it executable by the computer will be described. explain.

As shown in FIG. 23A, the program can be recorded in advance on a hard disk 102 or a semiconductor memory 103 as a recording medium built in the computer 101.

Alternatively, the program is executed as shown in FIG.
As shown in (B), the floppy disk 111 and the CD-R
OM (Compact Disc Read Only Memory) 112, MO (Magnet
o optical) disc 113, DVD (Digital Versatile Di)
sc) 114, magnetic disk 115, semiconductor memory 116
Can be temporarily (permanently) stored (recorded) in a recording medium such as. Such a recording medium can be provided as so-called package software.

The program is installed in the computer from the recording medium as described above,
As shown in (C), from the download site 121,
The data is wirelessly transferred to the computer 101 via an artificial satellite 122 for digital satellite broadcasting, or a LAN (Local Area N
etwork), the Internet 131
Is transferred to the computer 101 via a cable, and the computer 101
And so on.

Here, in the present specification, processing steps for writing a program for causing a computer to perform various kinds of processing do not necessarily have to be processed in chronological order in the order described in the flowchart, and may be performed in parallel. Alternatively, it also includes processing executed individually (for example, parallel processing or processing by an object).

The program may be processed by one computer, or may be processed by a plurality of computers in a distributed manner. Further, the program may be transferred to a remote computer and executed.

Next, FIG. 24 shows the computer 1 of FIG.
1 shows a configuration example.

As shown in FIG. 24, the computer 101 has a built-in CPU (Central Processing Unit) 142. An input / output interface 145 is connected to the CPU 142 via a bus 141, and the CPU 142 is connected to the CPU 142 by the user via the input / output interface 145.
When a command is input by operating an input unit 147 including a keyboard, a mouse, and the like, a ROM corresponding to the semiconductor memory 103 in FIG.
(Read Only Memory) The program stored in 143 is executed. Alternatively, the CPU 142 may execute a program stored in the hard disk 102,
Alternatively, the data is transferred from the network 131 and the communication unit 14
8, the program installed on the hard disk 102 or the floppy disk 111, the CD-ROM 112, and the MO disk 11 mounted on the drive 149.
3. A program read from the DVD 114 or the magnetic disk 115 and installed on the hard disk 102 is loaded into a RAM (Random Access Memory) 144 and executed. Then, the CPU 142 transmits the processing result to the LC via the input / output interface 145, for example.
An output unit 146 including a D (Liquid CryStal Display), a speaker, and the like is output as needed.

In this embodiment, speech is recognized,
Although the speech recognition result was translated, the present invention
The present invention is also applicable to the case where a sentence input by operating a keyboard or the like is translated.

In the present embodiment, the translation result is output as a synthesized sound. However, the present invention can be applied to a case where the translation result is displayed as a text.

Furthermore, the present invention can be applied to translations between Japanese and English and languages other than English and Japanese.

[0219]

According to the translating apparatus, the translating method, and the recording medium of the present invention, the correspondence for translating the input sentence into the translated sentence is at least one of the first and second languages. Based on the table described with the information, the input sentence is translated into a translated sentence corresponding to the input sentence. Therefore, high-precision translation is performed,
It is possible to output a translated sentence with a synthetic sound that does not cause discomfort.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an example of a conventional speech translation system.

FIG. 2 is a block diagram illustrating a configuration example of an embodiment of a speech translation system to which the present invention has been applied.

FIG. 3 is a plan view showing an example of an external configuration of a speech translation system to which the present invention is applied.

FIG. 4 is a block diagram illustrating a configuration example of a voice recognition unit 1;

FIG. 5 is a block diagram illustrating a configuration example of a machine translation unit 2.

FIG. 6 is a diagram showing language conversion data.

FIG. 7 is a block diagram illustrating a configuration example of a speech synthesis unit 3.

FIG. 8 is a flowchart illustrating the operation of the speech translation system in FIG. 2;

FIG. 9 is a diagram showing a word graph.

FIG. 10 is a view following FIG. 9;

FIG. 11 is a diagram showing prosody data and prosody tags.

FIG. 12 is a diagram for explaining the operation of the speech translation system in FIG. 2;

FIG. 13 is a plan view for explaining the operation of the speech translation system in FIG. 3;

FIG. 14 is a plan view for explaining the operation of the speech translation system in FIG. 3;

FIG. 15 is a plan view for explaining the operation of the speech translation system in FIG. 3;

FIG. 16 is a plan view for explaining the operation of the speech translation system in FIG. 3;

FIG. 17 is a diagram for explaining exchanges among the speech recognition unit 1, the machine translation unit 2, and the speech synthesis unit 3.

FIG. 18 is a flowchart for explaining the operation of the speech recognition unit 1.

FIG. 19 is a flowchart for explaining the operation of the machine translation unit 2;

FIG. 20 is a flowchart for explaining processing in steps S23 to S25 in FIG. 19;

FIG. 21 is a flowchart for explaining the operation of the speech synthesizer 3;

FIG. 22 is a flowchart for explaining processing in steps S62 and S63 in FIG. 21;

FIG. 23 is a diagram illustrating a recording medium to which the present invention has been applied.

24 is a block diagram illustrating a configuration example of a computer 101 in FIG.

[Explanation of symbols]

1 voice recognition unit, 2 machine translation unit, 3 voice synthesis unit, 4 display unit, 5 speaker, 6 operation unit, 6
A cursor key, 6B decision key, 6C cancel key, 7 control unit, 11 microphone (microphone), 12 AD conversion unit, 13 feature extraction unit, 14
Buffer section, 14A audio data buffer, 14
B feature buffer, 15 matching unit, 16
Acoustic model database, 17 dictionary database,
18 grammar database, 19 prosody information extraction unit, 21 text analysis unit, 22 language conversion unit,
23 text generator, 24 dictionary database,
25 grammar database for analysis, 26 language conversion database, 27 dictionary database, 28 grammar database for generation, 31 text analyzer, 32
Rule synthesis unit, 33 DA conversion unit, 34 dictionary database, 35 grammar database for analysis, 36 phoneme segment database, 101 computer, 102
Hard disk, 103 semiconductor memory, 111
Floppy disk, 112 CD-ROM, 113 MO
Disk, 114 DVD, 115 magnetic disk,
116 semiconductor memory, 121 download site, 122 satellite, 131 network, 141
Bus, 142 CPU, 143 ROM, 144 RA
M, 145 input / output interface, 146 output section, 147 input section, 148 communication section, 149 drive

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification code FI Theme coat ゛ (Reference) G10L 3/00 551C (72) Inventor Hideki Kishi 6-7-35 Kita Shinagawa, Shinagawa-ku, Tokyo Sony Stock In-house (72) Inventor Koji Asano 6-7-35 Kita-Shinagawa, Shinagawa-ku, Tokyo F-term in Sony Corporation (reference) 5B091 AA05 AA06 AA15 BA03 CA21 CB12 CB32 CC01 EA00 5D015 CC13 CC14 KK02 KK04 5D045 AA07 AB03 9A001 HH14 HZ17 HZ18

Claims (10)

[Claims] 1. A translation device for translating an input sentence in a first language and outputting a translated sentence in a second language, comprising: input means for inputting the input sentence; The input sentence is converted into the translated sentence corresponding to the input sentence based on a table in which the correspondence for translating the input sentence is described together with the prosody information of at least one of the first and second languages. A translation device, comprising: a translation unit that translates; and an output unit that outputs the translated sentence. 2. The translation according to claim 1, wherein said input means includes a voice recognition means for recognizing a voice, and outputs a result of said voice recognition by said voice recognition means as said input sentence. apparatus. 3. The voice recognition means includes extraction means for extracting prosody information of the voice, outputs a voice recognition result of the voice together with the prosody information of the voice, and the translator includes the voice recognition 3. The apparatus according to claim 2, wherein the result is translated using the prosody information. 4. The apparatus according to claim 1, further comprising a request unit for requesting the process information to the voice recognition unit, wherein the voice recognition unit outputs the process information when requested by the request unit. The translation device according to claim 3, wherein 5. The translation according to claim 1, wherein the output unit includes a speech synthesis unit that generates a synthesized speech corresponding to the translated sentence, and outputs the translated sentence as a synthesized speech. apparatus. 6. The translation means outputs the translated sentence together with its prosody information, and the speech synthesis means outputs a synthesized sound corresponding to the translated sentence,
The translation device according to claim 5, wherein the translation is generated using the prosody information. 7. The voice synthesizing means includes a requesting means for requesting the prosody information to the translating means, and the translating means transmits the prosody information when requested by the requesting means. The translation device according to claim 6, wherein the translation is output. 8. The apparatus according to claim 1, further comprising a storage unit that stores the table. 9. A translation method for translating an input sentence in a first language and outputting a translated sentence in a second language, comprising: an inputting step of inputting the input sentence; The input sentence is converted into the translated sentence corresponding to the input sentence based on a table in which the correspondence for translating the input sentence is described together with the prosody information of at least one of the first and second languages. A translation method, comprising: a translation step of translating; and an output step of outputting the translation sentence. 10. A recording medium storing a program for causing a computer to perform a translation process of translating an input sentence in a first language and outputting a translated sentence in a second language, wherein the input sentence is An input step of inputting, and a correspondence relationship for translating the input sentence into the translated sentence is based on a table in which prosody information of at least one of the first or second language is described. A recording medium recording a program including a translation step of translating an input sentence into the translation sentence corresponding to the input sentence, and an output step of outputting the translated sentence.