JP2006030384A - Text-to-speech synthesizer and text-to-speech synthesis method - Google Patents

Abstract

Provided is a text-to-speech synthesizer and a text-to-speech synthesis method capable of synthesizing speech by sufficiently expressing the characteristics of the language even when input text in which a plurality of languages are mixed is input. To do.
A language included in an input text in which a plurality of languages are mixed is determined, and the input text is divided for each determined language. The text dividing unit 2 outputs the divided text parts to the speech synthesis units 3 _{1 to} 3 _n provided for each language. The speech waveforms synthesized by the text speech synthesis units 31 to 3n are integrated by the speech waveform integration unit 4.
[Selection] Figure 1

Description

  The present invention relates to a text-to-speech synthesizer and a text-to-speech synthesis method that can process a plurality of languages.

  Text-to-speech synthesis is to mechanically synthesize speech based on a human speech generation mechanism based on input text obtained by typing. FIG. 9 shows a configuration of a conventional text-to-speech synthesizer 100. A text-to-speech synthesizer 100 includes a speech symbol string generation unit 101 that converts input text into a phonetic symbol string, a prosody generation unit 102 that converts a phonetic symbol string into prosodic data, and a waveform generation unit that generates a speech waveform from prosodic data 103.

  The phonetic symbol string generation unit 101 divides the input text into morphemes that are the smallest meaningful language units, and generates a phonetic symbol string with reference to a dictionary that stores information such as morpheme notation, part of speech, and reading. . This phonetic symbol string is a representation of input text using phonetic symbols, accent symbols, pause symbols, tone symbols, and the like.

  Then, the prosody generation unit 102 analyzes the input phonetic symbol string and determines the duration, fundamental frequency, and power for each phoneme. Information on each determined phoneme is output to the waveform generation unit 103 as prosodic data. Here, a phoneme is a unit of sound used in a certain language, and is the smallest unit that causes a difference in meaning.

  The prosody data output from the prosody generation unit 102 is divided into phoneme string information, phoneme time information, and pitch pattern information by the waveform generation unit 103. The phoneme string information includes a phoneme string, and the acoustic feature parameter corresponding to the selected phoneme string is expanded and contracted on the time axis based on the phoneme time information, and the pitch is changed based on the pitch pattern information. Converted to speech waveform.

  Through the processing as described above, the speech synthesizer 100 converts the input text obtained by type input or the like into a speech waveform.

JP 2001-14305 A

  By the way, a text-to-speech synthesizer is known that can process a plurality of languages such as Japanese and English (see, for example, Patent Document 1). For example, the electronic document processing apparatus described in Patent Document 1 processes a plurality of languages by selecting a speech synthesis engine based on attribute information indicating a language describing the electronic document.

  However, Patent Document 1 does not describe, for example, a method for selecting a speech synthesis engine when a plurality of languages are mixed in one input text. In the technique described in Patent Document 1, a plurality of input texts are included in one input text. In the case of mixed languages, speech synthesis could not be performed with the prosody of each language.

  The present invention has been made in view of such problems, and even when an input text in which a plurality of languages are mixed is input, it is possible to synthesize speech by sufficiently expressing the characteristics of the language. An object is to provide a text-to-speech synthesizer and a text-to-speech synthesis method.

  In order to achieve the above-described object, a text-to-speech synthesizer according to the present invention is a text-to-speech synthesizer that processes input text having a text portion of two or more languages, and includes a language included in the input text. Integrating the language discriminating means for determining and dividing into text parts for each language, a plurality of speech synthesizing means provided for each language for converting the text parts into speech waveforms, and the speech waveforms converted for each language Voice language integration means, and the language discrimination means outputs the text portion to the plurality of voice synthesis means according to the language of the text portion.

  The text-to-speech synthesizer according to the present invention is a text-to-speech synthesizer that processes input text having a text portion of two or more languages, the determination means for determining the language included in the input text, A language processing means for converting the determined text portion of each language into a phonetic symbol string; a plurality of prosody generating means provided for each language for converting the phonetic symbol string into prosodic data; and A speech waveform generation means for converting the speech symbol string to the plurality of prosody generation means according to the phonetic symbol string.

  The text-to-speech synthesis method according to the present invention is a text-to-speech synthesis method for processing input text having a text portion of two or more languages, determining a language included in the input text, and determining the text for each language. A language discriminating step for dividing the text into a portion, a plurality of speech synthesizing steps provided for each language for converting the text portion into a speech waveform, and a speech waveform integrating step for integrating the speech waveforms converted for each language. And the language discrimination step outputs the text portion to the plurality of speech synthesis steps according to the language of the text portion.

  The text-to-speech synthesis method according to the present invention is a text-to-speech synthesis method for processing input text having a text portion of two or more languages, the determination step for determining a language included in the input text, A language processing step for converting the determined text portion for each language into a phonetic symbol string, a plurality of prosody generation steps for each language for converting the phonetic symbol sequence into prosodic data, and the prosody data as a voice waveform A speech waveform generation step of converting the speech symbol sequence to the plurality of prosody generation steps according to the phonetic symbol sequence.

  The present invention determines a language included in an input text having a text portion of two or more languages, and outputs the text portion to the speech synthesizer of each language according to the language of the text portion. It can synthesize speech with prosody.

  Further, the language included in the input text having the text portions of two or more languages is determined, the text portion for each language is converted into a phonetic symbol sequence, and the phonetic symbol sequence is converted into a prosody of each language according to the phonetic symbol sequence. By outputting to the generating means, it is possible to synthesize speech with prosody of each language.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. A text-to-speech synthesizer shown as a specific example of the present invention determines a language included in an input text in which a plurality of languages are mixed, and synthesizes speech by sufficiently expressing the characteristics of the language.

(First embodiment)
FIG. 1 shows the overall configuration of a text-to-speech synthesizer 10 according to the first embodiment. The text-to-speech synthesizer 10 includes a language determination unit 1 that determines a language included in an input text in which a plurality of languages are mixed, a text division unit 2 that divides the input text for each determined language, and a divided text portion A text-to-speech synthesizer 3 _{1 to} 3 _n provided for each language for synthesizing a speech waveform, and a speech waveform integration unit 4 to integrate the speech waveforms synthesized by the text-to-speech synthesizers 3 _{1 to} 3 _n. Configured.

The language determination unit 1 determines a language included in an input text having a text portion of two or more languages. Moreover, the language determination part 1 provides the tag which shows language information for every determined language. For example, the tagged text for a Japanese and English mixed language sentence “This is dictionary” is described as follows.
<lang = japanese> This is </ lang><lang = english> dictionary </ lang><lang = japanense></lang>
The character string enclosed in <> is a tag, and the rest are text parts. <Lang = japanese> indicates that Japanese (Japanese) starts within the sentence, and </ lang> indicates that the language ends.

The language is determined by the following method, for example.
1. Judgment of language by information given from outside
For example, when attribute information such as a tag is assigned to each word, the language is determined based on the language information included in the attribute information.
2. Determine language by character type
For example, hiragana, katakana and kanji are judged as Japanese, and alphabets are judged as English.
3. Determine language by referring to a dictionary
For example, the language is determined by collating the input text with a word stored in the dictionary.
4). Determine language by character code
The language is determined by analyzing a character code that is a list of numerical values determined by the type of language.

The text dividing unit 2 cuts the input text into text portions to which tags are assigned for each language, and outputs the text portions to the text-to-speech synthesis units 3 _{1 to} 3 _n of the respective languages based on the language information of the tags.

The speech waveform integration unit 4 integrates the speech waveforms synthesized by the text speech synthesis units 3 _{1 to} 3 _n of each language. The speech waveform integration unit 4 transmits a waveform arrival notification to the text division unit 2 each time a speech waveform is input, and receives an end command from the text division unit 2, thereby integrating the voice waveform integrated so far. Is output.

  Next, the operation of the speech synthesizer 10 will be described with reference to the flowchart shown in FIG. First, the language determination unit 1 determines a language included in the input text (step S1), and assigns a tag indicating language information to each determined language to be divided into text parts for each language.

The tagged text is cut for each language by the text dividing unit 2 (step S2). The text dividing unit 2 first outputs the first text portion of the input text to the text-to-speech synthesis units 3 _{1 to} 3 _n of the language according to the language information of the tag. The text-to-speech synthesizers 3 _{1 to} 3 _n convert the input text portion into a speech waveform (step S3) and output it to the speech waveform integration unit 4.

When the speech waveform integration unit 4 receives the speech waveform output from the text speech synthesis unit, the speech waveform integration unit 4 transmits a waveform arrival notification that the speech waveform has arrived to the text division unit 2. When the text division unit 2 receives the waveform arrival notification from the waveform integration unit 4, the text division unit 2 outputs the second text portion from the beginning of the input text to the text-to-speech synthesis unit n of the language according to the language information. Similarly, the second text is also converted into a speech waveform by the text-to-speech synthesis units 3 _{1 to} 3 _n and sent to the speech waveform integration unit 4.

The speech waveform integration unit 4 integrates the speech waveforms output from the text speech synthesis units 3 _{1 to} 3 _n in the order in which they are input to the speech waveform integration unit 4 (step S4), and signals that the speech waveform has arrived. Send to text divider 2. Similarly, the third and subsequent text portions are processed until the text portion of the input text is completed (step S5).

  When the waveform arrival notification of the final text portion is received, the text dividing unit 2 transmits an end command to the speech waveform integrating unit 4. When receiving the end command, the speech waveform integration unit 4 outputs the speech waveform combined so far (step S6).

As described above, the text-to-speech synthesizer 10 according to the _first embodiment outputs two or more text parts to the speech synthesizers 3 _{1 to} 3 _n of each language according to the language of the text part. Even an input text having a text portion of a language can be synthesized with the prosody of each language.

(Second Embodiment)
FIG. 3 shows the configuration of the text-to-speech synthesizer 20 in the second embodiment. The text-to-speech synthesizer 20 includes a language determination unit 21 that determines a language included in an input text in which a plurality of languages are mixed, a speech symbol string generation unit 22 that converts a text part for each word into a phonetic symbol string, a change processing unit 23 for outputting a symbol string to the prosody generation unit 24 ₁ to 24 _n provided for each language, and the prosody generation unit 24 ₁ to 24 _n for each language to be converted in the prosodic data the voice symbol sequence, each prosody The prosody connection unit 25 connects prosody data from the generation units 24 _{1 to} 24 _n and the waveform generation unit 26 generates a speech waveform based on the prosody data.

Similar to the language determination unit 1 in the first embodiment, the language determination unit 21 determines a language included in an input text having a text portion of two or more languages. Moreover, the language determination part 1 provides the tag which shows language information for every determined language. For example, the tagged text for a Japanese and English mixed language sentence “This is dictionary” is described as follows.
<lang = japanese> This is </ lang><lang = english> dictionary </ lang><lang = japanense></lang>
The character string enclosed in <> is a tag, and the rest are text parts. <Lang = japanese> indicates that Japanese (Japanese) starts within the sentence, and </ lang> indicates that the language ends. Moreover, although it demonstrates using the tag in which linguistic information was described, embodiment of this invention is not restricted to this.

The language is determined by the following method, for example.
1. Judgment of language by information given from outside
For example, when attribute information such as a tag is assigned to each word, the language is determined based on the language information included in the attribute information.
2. Determine the language by the type of characters
For example, hiragana, katakana and kanji are judged as Japanese, and alphabets are judged as English.
3. Determine language by referring to a dictionary
For example, the language is determined by collating the input text with a word stored in the dictionary.
4). Determine language by character code
The language is determined by analyzing a character code that is a list of numerical values determined by the type of language.

  The phonetic symbol string generation unit 22 analyzes the input text sentence as shown in FIG. 4 to generate morpheme information, a prosody information generation unit 222 that generates phonetic symbols based on the morpheme information, and It is comprised. Here, a morpheme is a minimum unit of a meaningful character string such as a stem, a prefix, and a suffix, and is slightly smaller than a word. The phonetic symbol string is expressed by phonetic symbols or accent symbols.

  The text analysis unit 221 has a text analysis rule 223 and a dictionary 224. The text analysis rule 223 stores rules relating to morpheme arrangement (grammar, connection matrix, morpheme N-gram, etc.) and rules for assigning necessary information to morphemes not registered in the dictionary 224. The dictionary 224 stores information on registered morphemes such as notation, part of speech, and reading.

  The prosodic information generation unit 222 has a prosodic information generation rule 225 for analyzing morphological information and obtaining prosodic information. In the prosodic information generation rule 225, rules for performing phrasing and reading changes indicating phrase separation based on morpheme information are stored. Further, for each input morpheme, the dictionary 224 used in the text analysis unit 221 is also connected to obtain more detailed information for prosody generation. The information for prosody generation is, for example, information on how reading changes when a plurality of morphemes are combined to form a compound word, information on the movement pattern of the accent nucleus position, etc. .

  Note that the configuration of the phonetic symbol string generation unit 22 shown in FIG. 4 has a configuration in which information about morphemes and information for prosody generation are stored in one dictionary 224. Good. In other words, the text analysis unit 221 is connected to a dictionary that stores only information about morphemes, and the prosody information generation unit 222 is connected to a dictionary that stores only information for prosody generation. It can be performed. In addition, the text analysis rule 223, the dictionary 224, and the prosody information generation rule 225 store rules for each language, but may be configured to have rules independently for each language.

  Here, processing of input text will be described. The input text may be a single sentence or a plurality of sentences.

  The input text is divided into text portions for each language by the language determination unit 21. The segmented text portion is divided into morphemes by the text analysis unit 221 using the text analysis rules 223 and the dictionary 24, and information on each morpheme is obtained from the dictionary 224. For morphemes that are not registered in the dictionary 224, necessary morpheme information is generated using the text analysis rule 223. The morpheme information obtained in this way is sent to the prosodic information generation unit 222.

  The prosodic information generation unit 222 performs processing of adding prosodic information to the input sentence using the prosodic information generation rule 225 based on the morpheme information received from the text analysis unit 221. The prosodic information differs depending on the language, but in the case of Japanese, it is information such as an accent nucleus position, an accent strength, a pose position, a pose length, and a change in reading. Note that prosody information varies depending on the design of the apparatus, and thus it is not necessary to include all of the above information and is not limited to the above information.

  The prosodic information obtained by the processing in the prosodic information generation unit 222 is expressed as a phonetic symbol string and is output to the prosody generation unit. A phonetic symbol string is a representation of input text using phonetic symbols, accent symbols, pause symbols, tone symbols, and the like.

These symbols may define their own symbols or adopt existing symbols. For example, pronunciation symbols such as IPA (International Phonetic Alphabet) and SAMPA (Speech Assessment Methods Phonetic Alphabet), accent symbols, and the like may be used. Alternatively, a pause symbol such as ToBI (Tone and Break Indices), a tone symbol, or the like may be used. In this embodiment, for the sake of convenience, a unique symbol referring to Roman characters for the Japanese portion and SAMPA for the English portion is used. For example, for the input text “This is dictionary”, the phonetic symbol string output from the phonetic symbol string generator 22 is described as follows, for example.
<lang = japanese> korewa </ lang><lang = english> dIkS @ neri: </ lang><lang = japanense> desu </ lang>
A character string enclosed in <> is a tag, and the other character strings are phonetic symbol strings. <Lang = japanese> indicates that Japanese (Japanese) starts within the sentence, and </ lang> indicates that the language ends. Moreover, although it demonstrates using the tag in which linguistic information was described, embodiment of this invention is not restricted to this.

The switching processing unit 23 selects prosody generation units 24 _{1 to} 24 _n that output a phonetic symbol string. For example, the output of the phonetic symbol sequence is switched to the prosody generation units 24 _{1 to} 24 _n according to the language information given to the phonetic symbol sequence. For example, for the input phonetic symbol string <lang = japanese> korewa </ lang><lang = english> dIkS @ neri: </ lang><lang = japanense> desu </ lang>, the switching processing unit 23 Output “korewa” and “desu” to the Japanese prosody generation unit and “dIkS @ neri:” to the English prosody generation unit.

FIG. 5 shows a configuration of one prosody generation unit 24 _n among the prosody generation units 24 _{1 to} 24 _n . The prosody generation unit 24 _n includes a duration determination unit 51 for determining the duration of each phoneme, a fundamental frequency determination unit 52 for determining the fundamental frequency of each phoneme, and a power for determining the volume of each phoneme. And a determination unit 53.

  The duration length determination unit 51 includes a phonetic symbol string analysis unit 54 and a rule application unit 55. The phonetic symbol string analysis unit 54 has an analysis rule 56 for analyzing the phonetic symbol string. The analysis rule 56 stores rules for acquiring, for example, an accent nucleus position, accent strength, pose position, pose length, and reading change from a phonetic symbol string.

  The rule application unit 55 has a generation rule 57 for determining the duration of each phoneme. The generation rule 57 is a rule for determining the duration of each phoneme, and stores a rule for determining in which phoneme environment the default duration is expanded or contracted. The phoneme default duration is also stored. Here, the phoneme environment indicates whether or not the phoneme is a vowel, what the preceding and following phonemes are, what number phoneme of the syllable, the accent nucleus position, and the like.

  The duration length determination unit 51 uses this generation rule 57 to determine the duration length of each phoneme. Information necessary for applying the generation rule 57 can be obtained by analyzing the input phonetic symbol string. Information on the determined duration length is transmitted to the fundamental frequency determination unit 32.

  The fundamental frequency determination unit 52 includes a phonetic symbol string analysis unit 58, a generation rule application unit 59, and a time expansion / contraction unit 60. The phonetic symbol string analysis unit 58 has an analysis rule 61 for analyzing the phonetic symbol string. The analysis rule 61 stores rules for acquiring, for example, an accent nucleus position, accent strength, pose position, pose length, and reading change from a phonetic symbol string.

  The rule application unit 59 has a generation rule 62 for generating a fundamental frequency pattern. The generation rule 62 stores the value of the fundamental frequency pattern for the tone type such as the accent type, and rules for selection and modification of those patterns.

  Here, various types of rule selection methods can be considered depending on the device. In this embodiment, the rule is based on the accent type of the word to which the phoneme belongs, the strength of the accent, the sentence structure such as dependency, etc. Determine the applicability of.

  The fundamental frequency pattern data included in the generation rule holds fundamental frequency values that span a plurality of phonemes, and the patterns are classified for each prediction factor such as accent type and tone so that they can be selected according to the rule.

  The phonetic symbol string analysis unit 58 obtains information necessary for applying the generation rule 62 based on the analysis rule 61 of the input phonetic symbol string. The rule application unit 59 selects and transforms an optimal pattern using the generation rule 62 from the obtained information. The time expansion / contraction unit 60 deforms the selected pattern in accordance with the duration length of each phoneme already determined by the duration length determination unit 51, and determines the fundamental frequency for each phoneme.

  The power determination unit 53 includes a phonetic symbol string analysis unit 63 and a rule application unit 64. The phonetic symbol string analysis unit 63 has an analysis rule 65 for analyzing the phonetic symbol string. The analysis rule 65 stores rules for acquiring, for example, an accent nucleus position, accent strength, pose position, pose length, reading change, and the like from a phonetic symbol string.

  The rule application unit 64 has a generation rule 64 for determining the power of each phoneme. The generation rule 66 is a rule for determining the power of each phoneme.

  The power determination unit 53 uses the generation rule 66 to determine the power of each phoneme. Information necessary for applying the generation rule 66 can be obtained by analyzing the input phonetic symbol string. Information on the determined power is transmitted to the output generation unit 67.

  The duration information for each phoneme determined by the duration determination unit 51, the fundamental frequency for each phoneme determined by the fundamental frequency determination unit 52, and the power information for each phoneme determined by the power determination unit 53 are prosodic data. Are generated by the output generation unit 62 and output from the prosody generation unit 24n.

The prosody connection unit 25 arranges and connects the prosody data output from the prosody generation units 24 _{1 to} 24 _n so as to meet the order of the phonetic symbol strings input to the switching processing unit 23. When the prosodic data arrives, the prosodic connection unit 25 transmits an arrival notification to the switching processing unit 23. Since the switching processing unit 23 receives the arrival notification and outputs the next phonetic symbol string, the prosodic data is integrated in the order of the input text. In addition, when the arrival notification of the last phonetic symbol string is received, the switching processing unit 23 transmits an end notification to the prosodic connection unit 25. The prosody connection unit 25 receives the end notification and outputs the prosody data to the waveform generation unit 26.

FIG. 6 shows the configuration of the waveform generator 26. The waveform generation unit 26 generates a speech waveform from the prosodic data output from the prosody generation units 24 _{1 to} 24 _n . The waveform generation unit 26 includes a prosody data distribution unit 261 that distributes input prosodic data for each information, a segment selection unit 262 that generates an acoustic feature parameter of a phoneme string, and a parameter correction unit 263 that corrects the acoustic feature parameter. And a waveform assembly unit 264 that synthesizes the audio signal waveform while changing the pitch. The element selection unit 262 has audio data 265 that stores parameters indicating acoustic features.

  The prosody data distribution unit 261 divides the input prosody data into phoneme string information, phoneme time length information, and pitch pattern information, and outputs them to the segment selection unit 262, the parameter correction unit 263, and the waveform assembly unit 264, respectively.

  The segment selection unit 262 selects the phoneme sequence included in the phoneme sequence information with reference to the audio data 265 based on the input phoneme sequence information, and sets the acoustic feature parameter corresponding to the selected phoneme sequence as the audio data. Sequentially read from H.265 and output.

  The voice data 265 is the same as that used in the existing regular voice synthesizer, and is, for example, a parameter indicating the acoustic characteristics of voice such as a cepstrum coefficient. In addition, the length of each element is not fixed to a synthesis unit in CV, CVC (C: consonant, V: vowel), a unit corresponding to another synthesis unit, or a corpus-based synthesis method.

  The parameter correction unit 263 expands / contracts the acoustic feature parameters arranged on the time axis by the phoneme string on the time axis so as to be equal to the length of each phoneme by the phoneme time information input from the prosody data distribution unit 261. To do. In addition, the acoustic feature parameter is corrected in order to avoid inconsistency of the acoustic feature parameter at the connecting portion of the piece.

  The waveform assembling unit 264 synthesizes the audio signal waveform while changing the pitch based on the series of acoustic feature parameters output from the parameter correcting unit 263 and based on the pitch pattern information from the prosody data distributing unit 261. Output.

The text-to-speech synthesizer 20 having such a configuration outputs the text part to the prosodic generation units 24 _{1 to} 24 _n of each language in accordance with the language information of the phonetic symbol string, whereby the text parts of two or more languages are output. Can be synthesized with the prosody of each language.

In addition, the text-to-speech synthesizer 20 is configured to have prosody generation units 24 _{1 to} 24 _n for each language, but each rule may have a rule for each language in one prosody generation unit. In this case, speech synthesis can be performed with the prosody of each language without the configuration of the switching processing unit 23 and the prosody connection unit 25.

(Third embodiment)
FIG. 7 shows the configuration of the text-to-speech synthesizer 30 in the third embodiment. The text-to-speech synthesizer 30 includes a language determination unit 21 that determines a language included in an input text in which a plurality of languages are mixed, a speech symbol string generation unit 22 that converts a text part for each word into a phonetic symbol string, A switching processing unit 31 for switching output of the phonetic symbol string to the prosody generation units 24 _{1 to} 24 _n provided for each language according to the language determined by the determining unit 32, and a language for converting the phonetic symbol string into prosodic data Each prosody generation unit 24 _{1 to} 24 _n , a prosody connection unit 25 that connects the prosody data from each prosody generation unit 24 _{1 to} 24 _n, and a waveform generation unit 26 that generates a speech waveform based on the prosody data It is configured.

That is, the text-to-speech synthesizer 30 according to the third embodiment further includes a language determining unit 32 in the configuration of the text-to-speech synthesizer 20 according to the second embodiment. The output of the phonetic symbol string is switched to each of the prosody generation units 24 _{1 to} 24 _n according to the determined language. In addition, the same instruction | indication code | symbol is attached | subjected to the part corresponding to each part of the structure shown in FIG. 3 mentioned above.

  The language determination unit 32 determines the language that is the base of the phonetic symbol string input to the switching processing unit 31. A base language is like the native language of a speaker used to generate prosody as if a single speaker was speaking.

  Next, a language determination method in the language determination unit 32 will be described. Here, the phonetic symbol string <lang = japanese> korewa </ lang> <lang = english> dIkS @ neri: </ lang> <lang = japanense> desu </ lang> of the input text "This is a dictionary." Will be described. A character string enclosed in <> is a tag, and the other character strings are phonetic symbol strings. <Lang = japanese> indicates that Japanese (Japanese) starts within the sentence, and </ lang> indicates that the language ends. Moreover, although it demonstrates using the tag in which linguistic information was described, embodiment of this invention is not restricted to this.

1. The language assigned to the longest interval in the input phonetic symbol string
In the above example, since “dIkS @ neri:” has the largest number of phonemes in the input phonetic symbol string, English is determined as the base language.
2. Language assigned to the largest number of intervals in the input phonetic symbol string
In the above example, since Japanese has two sections and English has one section, Japanese is determined as the base language.
3. The language assigned to the first occurrence of the input phonetic symbol string
In the above example, since the section <lang = japanense> korewa </ lang> that appears first is Japanese, Japanese is determined as the base language.
4). The language assigned to the last occurrence in the input phonetic symbol string
In the above example, since the section <lang = japanense> desu </ lang> that appears last is Japanese, Japanese is determined as the base language.
5. The language assigned to the first and last appearing intervals in the input phonetic symbol string
In the above example, since the first appearing section and the last appearing section are in Japanese, Japanese is determined as the base language.

  Note that a language that appears in the input phonetic symbol string may be arbitrarily selected and used as a base language. Moreover, it is good also as a base language by having a base language list and selecting arbitrarily from the list. Further, a language designated from the outside may be used as the base language.

  When the base language cannot be determined by any one or more of the above methods, the prosodic control unit of the language corresponding to the language information of the tag can be used as in the second embodiment described above.

  As shown in FIG. 8, the switching processing unit 31 includes a reconversion unit 311 that reconverts a phonetic symbol string and a language conversion unit 312 that converts language information.

  The reconversion unit 311 reconverts the phonetic symbol string according to the language determined by the language determination unit 32. At that time, a conversion rule in which information about reading is stored is referred to. For example, when Japanese is determined as the base language, the phonetic symbol string is reconverted into a Japanese phonetic symbol string. In this case, for example, the phonetic symbol string <lang = english> dIkS @ neri: </ lang> is reconverted into <lang = english> dikushonarii </ lang>.

  The language conversion unit 312 changes the language information of the phonetic symbol string according to the language determined by the language determination unit 32. For example, when Japanese is determined as the base language, the language information given to the phonetic symbol string is Japanese. In this case, for example, the phonetic symbol string <lang = english> dIkS @ neri: </ lang> is converted to <lang = japanense> dIkS @ neri: </ lang>. That is, the language information of the output destination given to the phonetic symbol string is converted into the language determined by the language determination unit 32.

  As described above, by including the re-conversion unit 311 and the language conversion unit 312, it is possible to convert a phonetic symbol string of a language other than the base language into a phonetic symbol string of the base language and generate a prosody. Even if the input phonetic symbol sequence includes a phonetic symbol sequence other than the base language, the prosody can be generated as it is.

  For example, the phonetic string <lang = japanese> korewa </ lang> <lang = english> dIkS @ neri: </ lang> <lang = japanense> desu </ lang> for the input text "This is a dictionary." lang = english> korewa </ lang> <lang = english> dIkS @ neri: </ lang> By converting to <lang = english> desu </ lang>, only the pronunciation of the dictionary is fluent in English, and other parts Can express utterances that are in Japanese.

  Also, by converting to <lang = japanense> korewa </ lang> <lang = japanense> dikushonarii </ lang> <lang = japanense> desu </ lang>, you can express fluent Japanese utterances. .

  Alternatively, the language determination unit 32 may determine different base languages using different language determination methods, and convert the phonetic symbol strings and the language information thereof based on the different base languages. In this case, for example, the phonetic string <lang = japanese> korewa </ lang> <lang = english> dIkS @ neri: </ lang> <lang = japanense> desu </ lang Can be converted to <lang = english> korewa </ lang> <lang = english> dikushonarii </ lang> <lang = english> desu </ lang>, thus expressing a simple Japanese utterance be able to.

  Further, not only the base language but also the prosody of various languages can be specified, so that a unique utterance can be expressed.

  In the first to third embodiments, a tag indicating language information is assigned to each determined language. However, the present invention is not limited to this. For example, a different phonetic symbol string is used for each language. May be used.

1 is a block diagram showing a configuration of a text-to-speech synthesizer according to the present invention. It is a flowchart explaining operation | movement of the text-to-speech synthesizer concerning this invention. 1 is a block diagram showing a configuration of a text-to-speech synthesizer according to the present invention. It is a block diagram which shows the structure of an audio | voice symbol sequence production | generation part. It is a block diagram which shows the structure of a prosody generation part. It is a block diagram which shows the structure of a waveform generation part. 1 is a block diagram showing a configuration of a text-to-speech synthesizer according to the present invention. It is a block diagram which shows the structure of a switching process part. It is a block diagram which shows the structure of the conventional text-to-speech synthesizer.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Language determination part, 2 Text division | segmentation part, 3 _{1-3 n} text speech synthesizer, 4 Speech waveform integration part, 10 Text speech synthesizer, 20 Text speech synthesizer, 21 Language determination part, 22 Speech symbol string production | generation part, 23 switching processing unit, 24 _{1 to} 24 _n prosody generation unit, 25 prosody connection unit, 26 waveform generation unit, 30 text speech synthesis device, 31 switching processing unit, 32 language determination unit, 100 text speech synthesis device, 101 speech symbol string Generator, 102 prosody generator, 103 waveform generator

Claims (28)

A text-to-speech synthesizer that processes input text having text portions in two or more languages,
Language determination means for determining a language included in the input text and dividing the language into text portions for each language;
A plurality of speech synthesis means provided for each of the languages for converting the text portion into a speech waveform;
Voice waveform integration means for integrating the voice waveforms converted for each language,
The text speech synthesizing apparatus, wherein the language discrimination means outputs the text part to the plurality of speech synthesis means according to the language of the text part.   2. The text-to-speech synthesizer according to claim 1, wherein the language determining means determines a language based on attribute information previously given to the text portion.   2. The text-to-speech synthesizer according to claim 1, wherein the language discriminating unit judges a language with reference to a dictionary for identifying the text portion.   2. The text-to-speech synthesizer according to claim 1, wherein the language discriminating unit determines a language based on a character type and / or a character code of the text portion. A text-to-speech synthesizer that processes input text having text portions in two or more languages,
Determining means for determining a language included in the input text;
Language processing means for converting the determined text portion of each language into a phonetic symbol string;
A plurality of prosodic generation means provided for each of the languages for converting the phonetic symbol string into prosodic data;
A speech waveform generating means for converting the prosodic data into a speech waveform;
The speech processing apparatus according to claim 1, wherein the language processing unit outputs the phonetic symbol string to the plurality of prosody generation units according to the phonetic symbol sequence.   6. The text-to-speech synthesizer according to claim 5, wherein the determination means includes determining a language based on attribute information previously given to the text portion.   6. The text-to-speech synthesizer according to claim 5, wherein the determination unit determines a language with reference to a dictionary for identifying the text portion.   6. The text-to-speech synthesizer according to claim 5, wherein the determination unit determines a language based on a character type and / or a character code of the text part.   6. The text-to-speech synthesizer according to claim 5, wherein the language processing means further includes language determining means for determining a language as a base of the phonetic symbol string.   10. The text-to-speech synthesizer according to claim 9, wherein the language determining unit determines a language assigned to the longest section in the phonetic symbol string.   10. The text-to-speech synthesizer according to claim 9, wherein the language determining means determines a language assigned to the largest number of sections in the phonetic symbol string.   10. The text-to-speech synthesizer according to claim 9, wherein the language determining means determines a language assigned to a section that appears first and / or a section that appears last in the phonetic symbol string.   10. The text-to-speech synthesizer according to claim 9, wherein the language processing means includes re-conversion means for re-converting the phonetic symbol string according to the language determined by the language determination means.   10. The text-to-speech synthesizer according to claim 9, wherein the language processing means includes language conversion means for converting language information of the phonetic symbol string in accordance with the language determined by the language determination means. A text-to-speech synthesis method for processing input text having text parts in two or more languages,
A language determination step of determining a language included in the input text and dividing the language into text portions for each language;
A plurality of speech synthesis steps provided for each of the languages for converting the text portion into a speech waveform;
A speech waveform integration step of integrating the speech waveforms converted for each language,
The text speech synthesis method, wherein the language discrimination step outputs the text portion to the plurality of speech synthesis steps according to the language of the text portion.   16. The text-to-speech synthesis method according to claim 15, wherein the language determining step includes determining a language based on attribute information previously given to the text portion.   16. The text-to-speech synthesis method according to claim 15, wherein the language determination step determines a language with reference to a dictionary for identifying the text portion.   16. The text-to-speech synthesis method according to claim 15, wherein the language determination step determines a language based on a character type and / or a character code of the text portion. A text-to-speech synthesis method for processing input text having text parts in two or more languages,
A determination step of determining a language included in the input text;
A language processing step of converting the determined text portion of each language into a phonetic symbol string;
A plurality of prosodic generation steps provided for each language for converting the phonetic symbol string into prosodic data;
A speech waveform generation step of converting the prosodic data into a speech waveform;
The text speech synthesizing method, wherein the language processing step outputs the phonetic symbol sequence to the plurality of prosody generation steps according to the phonetic symbol sequence.   20. The text-to-speech synthesis method according to claim 19, wherein the determination step includes determining a language based on attribute information previously given to the text portion.   20. The text-to-speech synthesis method according to claim 19, wherein the determination step determines a language with reference to a dictionary for identifying the text portion.   20. The text-to-speech synthesis method according to claim 19, wherein the determination step determines a language based on a character type and / or a character code of the text portion.   20. The text-to-speech synthesis method according to claim 19, wherein the language processing step further includes a language determination step of determining a language as a base of the phonetic symbol string.   24. The text-to-speech synthesis method according to claim 23, wherein the language determining step determines a language assigned to the longest section in the phonetic symbol string.   24. The text-to-speech synthesis method according to claim 23, wherein the language determining step determines languages assigned to the largest number of sections in the phonetic symbol string.   24. The text-to-speech synthesis method according to claim 23, wherein the language determining step determines a language assigned to a section that appears first and / or a section that appears last in the phonetic symbol string.   24. The text-to-speech synthesis method according to claim 23, wherein the language processing step includes a reconversion step of reconverting the phonetic symbol string in accordance with the language determined in the language determination step. 24. The text-to-speech synthesis method according to claim 23, wherein the language processing step includes a language conversion step of converting language information of the phonetic symbol string according to the language determined in the language determination step.

Images