JP2002073081A - Voice recognition method and electronic equipment - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an English voice recognition method by which a rate of discrimination of alphabetic word is improved significantly with a simple configuration, and to provide electronic equipment of which operation convenience is improved by using a voice recognition technology. SOLUTION: In the English voice recognition method, the voice input of an alphabetic word is performed by substituting the word to alphabets that composes the word, such alphabets are inputted by substituting the alphabets to the combination of utterance of alphabetic characters of a plurality of languages, and the voice recognition is performed per above English alphabetic character. Such a voice recognition method is mounted on the electronic equipment.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition method and an electronic device, and more particularly to a technique effective for use in an electronic device such as a portable telephone for performing voice input.

[0002]

2. Description of the Related Art Speech recognition technology is a technology that allows a computer to directly recognize words and sentences spoken by humans, and has attracted attention as a simple input means that replaces keyboard input, pen input, etc. currently used as input means. . Examples of literature on speech recognition include Y.Obuchi, A.Koizumi, Y.Kit
ahara, J. Matsuda, and T. Tsukada, Proc. EUROSPEECH '
99, pp.2023-2026, 1999. Examples of dictation software are ViaVoice (IBM) and NaturallySpeaking.
(Dragon), AS as an example of voice recognition software
R1600 (L & H). IBM's "ViaV
oice "and Dragon's" Naturally Speaking "are primarily heavy-duty dictation software for Wintel PCs.

[0003]

In a digital electronic device such as a portable telephone, it is difficult to mount a keyboard having a large number of keys such as a personal computer because of the miniaturization thereof. At least 1
Since the two are small and densely arranged, they are inconvenient and impractical. Therefore, it is conceivable to use the above-mentioned speech recognition technology, which has attracted attention as an input means. Although text input has a very high level of performance due to its use, it requires a large-capacity memory and a high-performance CPU. Therefore, it is still not realistic to install such dictation software in a digital electronic device that needs to be small and consume low power, such as a mobile phone.

[0004] It is an object of the present invention to provide a speech recognition method which has a simple structure and which greatly improves the recognition rate of English words. Another object of the present invention is to provide an electronic device that improves usability by using a voice recognition technology.
The above and other objects and novel features of the present invention will become apparent from the description of the present specification and the accompanying drawings.

[0005]

The following is a brief description of an outline of a typical invention among the inventions disclosed in the present application. In the voice recognition method, a specific language is replaced with the alphabet constituting the same and the voice is input, and the alphabet is replaced with a combination of the utterances of the characters in a plurality of languages and the input is performed. Do.

The following is a brief description of an outline of another typical invention disclosed in the present application. In an electronic device that takes in and digitizes a voice signal in an input unit, extracts a feature of the voice signal input in a signal analysis unit, compares the feature with an acoustic model prepared in advance, and determines a character model, an input of a specific language is performed. Is input by replacing the alphabet with the constituent alphabet, and the input is performed by replacing the alphabet with a combination of utterances of characters in a plurality of languages, and the voice recognition by the voice recognition unit is performed in units of the characters of the alphabet.

[0007]

FIG. 1 is a block diagram showing an embodiment for explaining a speech recognition method according to the present invention. The method of English speech recognition in this embodiment is performed according to signal processing in each block in FIG. The uttered sound is first converted into a digital signal in a voice input unit. In this embodiment, in order to realize a high speech recognition rate with a simple configuration, it is necessary to input speech not in words or the like in a specific language but in units of the alphabet corresponding to the spelling. Features.

According to the Japanese dictionary, "letter" refers to each of the spellings that indicate pronunciation, such as "kana", "alphabet", and "sankanji". In English, "phoneme" is used. (Phony). This phoneme (phoneme) refers to a phoneme, which means the smallest phonetical unit of a language.

The above-mentioned spelling input in units of characters is performed by inserting a fixed silence period for each character, or by inserting a key input signal indicating a break of the character. In consideration of usability, it is considered useful to insert a silence period for each character. However, if a more accurate character separation is performed, it is useful to use a key signal.

[0010] The voice signal input from the voice input unit in character units is sent to a voice analysis unit, where a feature extraction process is performed. Specifically, it is a short-time frequency analysis. The analysis result is collated with an acoustic model prepared in advance in the collation unit, and a determination process is performed. The one with the highest score is displayed as the recognition result.

The acoustic model used in the matching unit is
Generally, the HMM, the word dictionary, and the context dictionary are connected. Here, the HMM is a so-called Hidden Markov Model, which is a reference model serving as a basis for recognition. A combination of the HMM and a context dictionary incorporating a word dictionary, context data, and the like is an acoustic model.

Although the HMM which is the basis of the above speech recognition is relatively light software, in order to perform speech recognition in a specific language corresponding to dictation as in the past, a word dictionary or context dictionary of the language must be used. It is essential to provide it, and it is likely to be heavy depending on the application.
In dictation softfare, which is required to recognize long sentences in real time, word dictionaries need to provide a large amount of data such as contexts and sentence examples.
A high-performance CPU for processing data at high speed and a large-capacity memory are required.

In a portable electronic device, it is necessary to reduce the number of components and suppress power consumption in order to enable long-term operation. Use as little memory as possible
Power is also limited. That is, it is required that the data amount including the dictionary be as light as possible. Therefore, in the English speech recognition method according to the present invention, by inputting an English word in units of characters and performing speech recognition in units of characters, basically the words are input as indicated by dotted lines in FIG. It greatly simplifies dictionaries and contest dictionaries.

In this embodiment, voice input by spelling is employed in order to maximize the input efficiency of English voice. For example, “butter” becomes “bi: ju: ti: ti: i:
a: r ", which is much more informative than simply saying" b * t * r "(patters), which greatly improves its recognition rate. It consists of 26 letters of the alphabet Is so much more lightweight that it can be entered and recognized exactly with software that is much lighter than anything that requires heavy equipment software that uses the phonetic symbols of every word and its accompanying audio data. You.

If a word can be input correctly, a sentence can be input similarly. This can be considered as an audio version of a word processor. That's exactly what an English word processor does, hitting "b", "u", "t", "t", "e", and "r" keys on the keyboard when typing "butter".

Another feature of the present invention is to use a Greek alphabet in addition to the above.
In the above “butter” example, “b”, “u”, “t”, “e”,
Use β (beita: heater), υ (ju: psilan: epsilon), τ (tau: tau), ε (eps * lan: epsilon), ρ (rou: low), etc. instead of “r” etc. That is. “Butter” to “bi: ju: t
When you say "i: ti: i: a: r", the output is “gutter”
Let's say That is, "b" and "g" are mixed. If you say “β” continuously at this time, the output automatically becomes the correct “butter”. When such crosstalk is expected in advance, “beita ju: ti: ti: i: a: r”
You can eliminate crosstalk by saying β (b)
In addition, δ (d), γ (g), π (p), τ (t) and the like are used. In addition, α (a), κ (k), ν (n), μ (m), ω
(o), ρ (r), σ (s), etc. can also be used as appropriate.

By adopting the above-described recognition method,
The recognition rate of words such as English can be increased to the utmost. That is, in spelling input, the syllables to be recognized are basically limited to alphabets, so that the recognition rate increases. Furthermore, the exclusion of alphabet combinations that do not correspond to words is also effective in increasing the recognition rate. For example, replace “butter” with “bi: ju: ti: ti:
When you say "i: a: r", you don't mistake the first "b" for the usually confusing "d" because there is no word "dutter".

Although the spelling input is effective by itself, the Greek spelling is used together to make the spelling input more effective. This is useful for further enhancing the differences between the alphabets. Taking English as an example, the so-called "e problem" can be avoided. That is, the ambiguity of “b”, “d”, “e”, “g”, “p”, “t”, etc. is expressed as “β”, “δ”, “ε”, “γ”, “π”,
It can be avoided by using “τ” or the like together. This is because English and Greek alphabets that are far apart from each other, that is, those that have little similarity can be selected. The Greek alphabet is not as familiar as English, but it is still relatively familiar and easy to remember.

The present invention has universality that it can be applied not only to English but also to European and American languages such as French, German, and Russian. That is, for example, by using a combination of French and Greek alphabets, the word recognition rate of French can be increased to the maximum.

When software or the like corresponding to the voice recognition method according to the present invention is installed in a portable electronic device, it is a matter of course that the effect of improving the efficiency of voice input by improving the word recognition rate is great. What is bigger than that is the sense of security that you can always enter. This is a huge thing from the standpoint of using the product, and it cannot be used if you can't get the correct input no matter how many times you say it. This is true regardless of the product to which the present invention is applied.

In the speech recognition method according to the present invention, several commands, such as "Capital letter" (uppercase), "hyphen" (hyphen), "comma" (comma), and "period" are input to the spelling input as described above. (period),
“Colon”, “space”, “new pa”
A command such as "ragraph" (line feed) may be added. By providing such a command, it is possible to easily input a sentence, and the effect is extremely large. Must be input in advance, and the recognition rate decreases remarkably as the number of words increases to 1,000 or 10,000. With the present invention, the recognition rate does not change even if the number of words increases. Can always be entered.

In the speech recognition method of this embodiment, since a small number of speeches, such as English and Gasha alphabets, are basically used, the data amount of the acoustic model can be reduced as much as possible. By combining the results of the alphabet characters, all types of words and sentences can be input as a result. Since the amount of data in the speech recognition can be reduced in this way, the central processing unit CPU for processing the data can be reduced in the power consumption of the RISC.
(Reduced instruction set computer) type can be used, and the memory capacity may be small. As a result, the speech recognition method according to the present invention is an optimal input method for a portable electronic device. Even if a specific control signal or operation command is performed by voice to improve the usability, the word dictionary and the context dictionary need only be small.

FIG. 2 is an external view of one embodiment in which the present invention is applied to a portable interpreter. The portable interpreter of this embodiment is directed to an English-Japanese interpreter. The method of using the portable interpreter of this embodiment is as follows. When the utterance button 1 is pressed and uttered toward the microphone 2, the recognition result is displayed on the display device 3. When a correct result is obtained, the user presses the OK button 4 to shift to sentence example search. Use the scroll button 5 to search and find the desired sentence.
Press to display the translation. When the utterance button 1 is further pressed, the voice of the translated sentence flows through the speaker 6.

An example in which the voice recognition method according to the present invention is applied to this portable interpreter will be described using "Thank" as an example. Voice button 1
Press and bring your mouth close to microphone 2 and “ti: eit * ei en ke
Say "i" (Tee H, I, E, N, Kay). The recognition result is displayed as "Thank" on the display device 3. At the time of recognition, the spell check is performed automatically, and words not in the dictionary are excluded. When the recognition result is checked on the display device 3 and the OK button 4 is pressed, a sentence example is displayed.

Select "Thank you for your help."
When the K button 4 is pressed, the translated sentence "Thank you. Then, when the utterance button 1 is pressed, a sound flows from the speaker 6. The selection example by sentence example search was described, but if it is a short sentence, "ti: eit * ei en kei, space,
ju: ". The display device 3
Displays "Thank you". The audio output is "thank you". Here, the pronunciation corresponding to the alphabet h
In eit *, the phonetic symbols are replaced with *.
The same applies to the phonetic symbols of the Greek alphabet. same as below.

Basically, you can input by spelling,
In some cases, it is difficult to input due to the surrounding noise or the like. In such a case, it is effective to use the Greek alphabet together. Take “thank” as an example, “th”,
For example, instead of “a”, “n”, and “k”, θ (sea), α (alf), ν (nyaw), κ (kappa) are used. It is also effective to use at the time of correction. In the same example, when the display of the result becomes “think”, the correct result “thank” can be easily obtained by replacing the “a” part with α and re-inputting.

Since the Greek alphabet corresponding to the English alphabet is not always present, the combination is fundamental. The combination of English and Greek alphabets not only avoids the "e problem", but also increases the distance between the alphabets, resulting in the ultimate increase in recognition rate. Here, "e
The "problem" refers to a problem in which similarities in utterances such as b, d, e, g, p, and t interfere with the recognition rate.

The portable interpreter shown in FIG. 2 can be used as it is as a portable word processor. The method of use is as follows. First say “Word Processor” as a command, then say “File” and “New”. A new entry screen is displayed. “Ti
"tle of the invention". The result is "Idol
of the convention ”is displayed. Move the pointer to“ I ”in“ Idol ”using the scroll buttons, and then touch“ ti: ai
ti: el i: "or" tau iota
tau lambda epsilon "(tau iota tau lambda epsilon).

Alternatively, a combination of English and Greek alphabets may be used. “Tau ai tau el i”. This gives the title "Title of a
ntion "is displayed. Then, move the pointer to" a "and say" θ eit * i: "(theta H e).
e ". Similarly, enter the correct input" Title of the
Invention "is obtained. Then, enter the text, enter the text, and correct it as necessary. Complete the input. Sometimes, you can use spelling in the middle or from the beginning. There are also technical terms, personal names, and place names that are not in the dictionary. This is especially useful when inputting, etc. When the whole sentence is completed, say “Store” to save the input text and finish the work.

In such a portable word processor,
Since both text-to-speech input and spelling input are used,
There is an effect that the input efficiency is improved. In addition, if spell input is adopted as needed, the input operation can be performed without worrying about the surroundings. Frequent use of the Greek alphabet can enhance this effect.

The portable interpreter shown in FIG. 2 can be directly replaced with a palmtop personal computer having a portable telephone function. For example, e-mail (e-ma
il) Another example of using the present invention for inputting a text sentence will be described below.

First, “e-mail” is used as an e-mail command.
To say. The screen switches to creating a new mail. Select an address from the address list displayed on the display screen.
Enter text. Say “How are you” “Question mark”. "How are you?" If it is necessary to correct the spelling, it is corrected by the Greco-Roman spelling input according to the present invention. Then, input the text of the mail in the same manner. Finally, say "Send mail" as a command to send the mail.

The above palmtop type personal computer may be a portable telephone, a notebook type personal computer connected to a network, or the like. Further, as a secondary effect of using the present invention, by appropriately changing the input to the spelling method, even if there is a person listening on the side, it can be used without concern.

The functions and effects obtained from the above embodiment are as follows. (1) In the speech recognition method, a specific language is replaced with an alphabet constituting the language and the speech is input, and the alphabet is replaced with a combination of utterances of the characters in a plurality of languages. The effect of this is that the identification rate can be significantly improved with a simple configuration.

(2) In addition to the above, the input of the characters in the above-mentioned plurality of languages can easily identify the characters having a high similarity by replacing a part of the English alphabet with the Greek alphabet, so that it is possible to perform one-time input. This has the effect of significantly improving the identification rate in recognition.

(3) In addition to the above, in the signal processing of the audio signal, a digital signal is converted into a digital signal in an input unit, a feature extracting process is performed in a voice analyzing unit, and an acoustic model including a hidden Markov model prepared in advance in a collating unit. The collation enables recognition by simple signal processing, so that an inexpensive memory or CPU can be used, and recognition by simple software can be achieved.

(4) An audio input unit for taking in an audio signal and digitizing it, and an audio signal processing unit for extracting features of the digitized audio signal and collating with a previously prepared acoustic model to determine a character model. The input of a specific language is replaced with the alphabet constituting the same, and the voice is input.Also, the alphabet is replaced with a combination of the utterances of the characters in a plurality of languages, and the input is performed. By using the combination of the character units, it is possible to obtain an electronic device in which the key operation is simple and characters and words can be easily input.

(5) In addition to the above, the input of the characters in the above-mentioned plurality of languages can be performed by replacing a part of the English alphabet with the Greek alphabet and uttering the voice. This leads to an effect that leads to a significant improvement.

Although the invention made by the inventor has been specifically described based on the embodiment, the invention of the present application is not limited to the embodiment, and various modifications can be made without departing from the gist of the invention. Needless to say. For example, it has universality that it can be applied to Western languages such as English, French, German, and Russian as described above. That is, for example, by using a combination of French and Greek alphabets, the word recognition rate of French can be increased to the maximum. An electronic device equipped with a voice recognition function using the voice recognition method according to the present invention can be similarly applied to a car navigation system and the like in addition to the above embodiments.

[0040]

The effects obtained by typical ones of the inventions disclosed in the present application will be briefly described as follows. In the voice recognition method, a specific language is replaced with the alphabet constituting the same and the voice is input, and the alphabet is replaced with a combination of the utterances of the characters in a plurality of languages and the input is performed. By doing so, the identification rate can be significantly improved with a simple configuration.

An audio input unit for receiving and digitizing an audio signal, and an audio signal processing unit for extracting features of the digitized audio signal and collating with a previously prepared acoustic model to determine a character model; The input of a specific language is replaced with the alphabet constituting the same and the voice is input, and the alphabet is replaced with a combination of the utterances of the characters in a plurality of languages, and the input is performed. By using the combination of the above, an electronic device can be obtained in which key operations are simple and characters and words can be easily input.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment for explaining a voice recognition method according to the present invention.

FIG. 2 is an external view showing an embodiment when the present invention is applied to a portable interpreter.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Voice button, 2 ... Display device, 3 ... Display screen, 4 ... OK
Buttons, 5 ... scroll buttons, 6 ... speakers.

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) H04M 1/00 G10L 3/00 537H 1/725 551A 551B (72) Inventor Makoto Tanaka Kodaira City, Tokyo 5-22-1, Honcho, Hitachi, Ltd., within SHI-LSI Systems (72) Inventor Hideki Uchidate, 5-22-1, Josui-Honmachi, Kodaira-shi, Tokyo F term (reference) 5D015 AA05 BB02 HH23 KK02 5K027 AA11 BB01 HH20

Claims (5)

[Claims] In a speech recognition method, a specific language is replaced by an alphabet constituting the same and speech input is performed. The alphabet is replaced by a combination of utterances of a plurality of languages and input. A speech recognition method characterized in that the method is performed in a combination of character units. 2. The speech recognition method according to claim 1, wherein the input of the characters in the plurality of languages is uttered by replacing a part of the English alphabet with a Greek alphabet. 3. The signal processing according to claim 1 or 2, wherein the signal processing of the audio signal is performed by converting the audio signal into a digital signal in an input unit, performing a feature extraction process in an audio analysis unit, and including a hidden Markov model prepared in advance in a matching unit. A speech recognition method including a method of determining a character base by matching with a model. 4. An audio input unit for inputting an audio signal, and an audio signal processing unit for digitizing the input audio signal, extracting features of the input audio signal, collating with a previously prepared acoustic model, and performing character discrimination. The input of a specific language is replaced with the alphabet constituting the same, and the input is made by voice. The alphabet is replaced with a combination of the utterances of the characters in a plurality of languages, and the input is performed. An electronic device characterized in that it is performed in a combination of units. 5. The electronic device according to claim 4, wherein the input of the characters in the plurality of languages is uttered by replacing a part of the English alphabet with a Greek alphabet.