JPH07302098A - Word voice recognition device - Google Patents

Abstract

PURPOSE:To make the word having a highest degree of similarity as a recognized word by comparing plural word data and the word data which are classified into basic vowels. CONSTITUTION:A vowel cutout section 4 receives word unit digital data constituted by the combination of sound syllables supplied by an A/D conversion section 2, cut outs the sound syllable of a last section (the last end), the sound syllable of a front section (the forefront) or the vowel of the last syllable and the front syllable and outputs them to a first stage voice recognition section 7. The section 7 compares the vowel section of the last section syllable of a cutout inputted word and the vowel comparison specimen which is beforehand registered in a vowel storage section 5 and the vowel having the highest degree of similarity comparison specimen is made as the recognition result. A second stage voice recognition section 8 reads the word selected by the section 7 as a comparison specimen, compares the specimen and the inputted word recorded in a backup memory 3 and the word having a high degree of similarity is outputted as a recognition result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a word voice recognition device, and more particularly to a word voice recognition device capable of highly accurate word voice recognition.

[0002]

2. Description of the Related Art A voice recognition device for recognizing a voice generated by humans is expected to be used in a wide variety of fields such as classification of packages by delivery destination and wireless control of various devices. On the other hand, if the number of words to be recognized increases, the processing time required for recognition will increase and the recognition rate will decrease, making it impractical. Therefore, the number of words to be recognized has a certain limit. There is.

Conventionally, in word speech recognition processing, the word speech is divided into syllables consisting of a combination of minimum units of vowels and consonants, and each syllable is united so that it is not easily affected by an increase in the number of target words. Many attempts have been made to perform recognition processing. Also, JP-A-59-160
No. 197 discloses recognition processing in syllable units composed of consonants and vowels. In this process, vowel recognition is performed by paying attention to the vowel of the syllable, and then syllable recognition processing including the vowel as a recognition result in the rear part is performed.

However, the above-mentioned method is a highly versatile word speech recognition method if its recognition accuracy is sufficiently high, but it is very difficult to divide the actually uttered word speech into syllable units. Yes, recognition in syllable units that are not correctly delimited does not make sense. Therefore, theoretically, the recognition processing method that performs the recognition processing for each word including the combination of the recognized syllables after the recognition processing for each syllable as described above, which is extremely versatile, has a problem that the configuration becomes complicated. Since it has, it has not been put to practical use.

Therefore, the following recognition processing has been put to practical use for word speech recognition in which the number of words to be recognized is large. That is, first, the words are divided into a plurality of groups in advance based on some predetermined associations,
A name is given to each group, the group name is uttered before the speech recognition processing of words, and the speech recognition processing is performed on the uttered group name. Next, only the words belonging to the group obtained as a recognition result are prepared as registered samples, that is, the words to be recognized, and subsequently, between the original recognized speech and the registered sample belonging to the group. Recognition is performed by performing comparison processing.

For example, when recognizing several hundred station names in the suburbs of Tokyo by voice, first, "Yamanote Line", "Chuo Line" and "Sobu Line"
Each station is divided into groups, such as "Saikyo Line" and "Nanbu Line". Next, say the group name to which the recognized station name belongs. For example, if the station name "Osaki" that belongs to the "Yamanote Line" is the recognized station name, first say "Yamanote Line", recognize that the input voice is "Yamanote Line", and then enter the "Yamanote Line" group. The station name to which it belongs becomes the target registration sample for the next original station name to be recognized. Next, when you say "Osaki", the station name (the number of which is limited) belonging to the "Yamanote Line" group is compared with the vocal station name, and the station name of the registered sample with the highest degree of similarity is recognized. And

[0007]

As described above, the practical use of the speech recognition apparatus demanded in many fields is limited at present because the number of words to be recognized is limited in consideration of the recognition rate. Then, various limits occur. If you try to recognize a large number of words by voice over these limits, you should divide them into groups and utter the group name as described above.
Since voice recognition, and subsequently utterance of the recognized word, recognition and two-stage utterance and recognition processing are required, it is troublesome and the user has to remember the group name. Furthermore, it may be difficult to associate places such as place names and person names in grouping.

SUMMARY OF THE INVENTION An object of the present invention is to provide a word voice recognition device capable of high-speed and highly accurate recognition processing of a large number of word voices with a relatively simple structure.

[0009]

In order to solve the above-mentioned problems, a word voice recognition apparatus of the present invention is provided with a vowel data storage means for storing data about a predetermined vowel and a plurality of word data for each word. A vowel group storing means for classifying and storing into a plurality of vowel groups according to the type of vowel contained in a predetermined syllable among the syllables that constitute the vowel, and a vowel containing the predetermined syllable for the input word speech as a basic vowel A vowel extracting means for extracting, comparing the data of the basic vowel extracted by the vowel extracting means and the data stored in the vowel data storing means, and recognizing the vowel with the highest degree of similarity as the basic vowel. 1 recognition means, the data of the input word voice, and the data of the words classified into the basic vowels recognized by the first recognition means are compared,
Second recognition means for outputting the word having the highest degree of similarity as a recognition word. Here, the predetermined syllable can be the last syllable, the frontmost syllable, or the two syllables of the frontmost part and the last part of the syllables forming the word.

[0010]

According to the present invention, the data of a plurality of words are classified into a plurality of vowel groups according to the type of vowel contained in a predetermined syllable of the syllables forming each word and stored. About the last syllable, the frontmost syllable or the two syllables of the frontmost part and the last part of the syllables that compose the word, the vowels included in the predetermined syllables are extracted as the basic vowels. A vowel having the highest degree of similarity is recognized as the basic vowel by comparing with the data, and the word having the highest degree of similarity is compared with the data of the words classified into the recognized basic vowels to be the recognized word.

[0011]

Embodiments of the present invention will now be described with reference to the drawings. In the word voice recognition apparatus according to the embodiment of the present invention, all the target words are first classified into a predetermined group in order to reduce the number of recognition target words in one voice recognition process. The classification method is as follows:
With the vowel part of the syllable at the defined position (for example, the last syllable of the word, the frontmost syllable, and these two syllables) as the key, "a", "i", "u", "e", "O",
Recognize "n" as a registered sample, and the recognition target words are "a" group, "i" group, "u" group,
The group to which the “E” group, the “O” group, and the “N” group belong is classified. After that, only the recognition target words in the recognized group are registered samples,
Perform recognition processing.

In this embodiment, since the vowel of a predetermined syllable in a word is focused on, it is not necessary to classify the recognition target words by some association and assign the group name in advance as in the prior art, and the recognition is not performed. There is no need to utter a group name before uttering the target word for recognition, and there is no need for the user to remember the group name.

That is, in the present embodiment, in the case of a speech recognition apparatus in which a large number of recognition target words are used as registered samples, the recognition unit at the first stage automatically classifies the recognition target words as vowel keys into groups, The recognition process is performed by using the recognition target words in the group identified by the recognition unit as the registered sample.

FIG. 1 is a block diagram showing the configuration of an embodiment of a word voice recognition apparatus according to the present invention. The voice signal input to the voice output unit 1 in word units is converted into digital data in the A / D conversion unit 2, and the backup memory 3 for temporarily storing the word unit data for voice recognition processing in the second stage.
Is supplied to the vowel extraction unit 4 which extracts the vowel of the final syllable of the word. The vowel storage unit (first group registered voice unit) 5 and the word storage unit (second group registered voice unit) 6 store in advance a comparative sample to be compared with the recognized word during voice recognition. The vowel storage unit 5 stores vowel comparison samples (“a”, “i”, “u”,
"E", "O", "N") is stored, and the word storage unit 6 stores
At the time of the second stage speech recognition, word comparison samples to be compared with the recognized word are classified into groups and stored.

The vowel cutout unit 4 receives word-by-word digital data composed of a combination of syllables supplied from the A / D conversion unit 2, and receives the last (final) syllable, the frontmost (first) syllable, Alternatively, the vowels of the last syllable and the frontmost syllable (in this example, the last syllable) are cut out, and the first-stage speech recognition unit 7
Output to.

The first-stage voice recognition unit 7 compares the vowel part of the last syllable of the input word cut out by the vowel cutout unit 4 with the vowel comparison sample registered in advance in the vowel storage unit 5,
The recognition result is the vowel that has the comparative sample with the highest similarity.

The second-stage voice recognition unit 8 is classified into the vowel group selected by the first-stage voice recognition unit 7 and reads a word registered in advance as a comparative sample, which is first recorded in the backup memory 3. The comparison process with the existing input word is executed, and the word of the comparative sample having the highest similarity is output as the recognition result.

The signal processing unit 9 outputs the recognition result thus obtained by the second-stage voice recognition unit 8 as a corresponding character control signal.

The vowel extraction of the last syllable forming a word in the vowel cutout unit 4 goes back from the last time when the threshold level Th is cut as shown in FIG. By extracting only the tn1 time portion of the input signal, sufficient vowel information that can be practically used can be obtained. According to the experiment, it was confirmed that the time tn1 should be set to 100 to 150 msec even in the case of early vocalization. In addition, even when uttering slowly, the length of the consonant part hardly changes and only the vowel part becomes long, and tn1 is 10 regardless of the speed of utterance.
It was also confirmed that it may be set to 0 to 150 msec.

The vowel extraction of the frontmost syllable of a word is shown in FIG.
As shown in (B), the cut-out time ts2 may be set after the delay time ts1 (usually about 30 msec) necessary for removing the consonant portion is passed from the first time point when the threshold level Th is exceeded. The vowel part does not cover the consonant part of the next syllable, and when it is desired to cut out for a long time but the vocalization is fast, the consonant + vowel part is shortened to about 200 msec, and the time ts2 is set to 100 to 150 msec.

As a concrete operation example of the present embodiment, the recognition processing of the word station name "Shibuya" with respect to the voice recognition of the station name in the suburbs of Tokyo will be described below.

In FIG. 1, from the voice output unit 1, "Shibuya"
When the word voice of is input through a microphone or the like, this input signal is converted into digital data by the A / D conversion unit 2 and then digital data corresponding to “Shibuya” is temporarily stored in the backup memory 3. It Then, in the vowel cutout unit 4, for example, the vowel of the last syllable of the word is cut out. In this way, the vowel "a" of the final syllable "ya" of the word "Shibuya" is cut out and supplied to the first stage speech recognition unit 7.

The first-stage voice recognition unit 7 has pre-registered "A", "I", "U", "E", "O", "N".
The voice recognition process is performed by using the data of (3) as a sample, and the recognized vowel is recognized as "A".

The second-stage voice recognition unit 8 selects the group "a" from the registered words registered in advance by grouping as shown in FIG. 3, and the registered word "Shinagawa" of the group "a" is selected. From 20 to “Yokohama” as the comparison sample data, the recognition processing is performed by comparing with the data of the word “Shibuya” previously stored in the backup memory 3.

As shown in FIG. 3, the sample of the “A” group registered in advance has only 20 words, and the word “Shibuya” having the highest degree of similarity is identified from this small sample, and the final sample is identified. The recognition result will be obtained. Therefore, as in the conventional case, the utterance of the group name and the utterance of the original word, such as the utterance of the original word, become uncertain, and the word can be recognized by the utterance of one time.

In the present embodiment, the grouping of words is performed based on the vowels included in the last syllable of each word, but may be performed based on the vowels included in the frontmost syllable of each word. , Can also be performed based on the vowels contained in both the frontmost and last syllables. When grouping is performed based on the vowels included in both syllables, the number of words included in each group is drastically reduced, which enables more highly accurate word recognition. For example, in the grouping based on the vowel of the last syllable, as shown in FIGS. 2 (A) and 2 (B), the number of words included in each vowel is about 20, while the number of words in both the front part and the last part is large. Grouping based on vowel syllables dramatically reduces the number of words belonging to each group, as shown in FIG.

[0027]

As described above, the word voice recognition apparatus of the present invention can recognize a word at high speed and with high accuracy by uttering a word to be recognized once.

[Brief description of drawings]

FIG. 1 is a configuration block diagram of an embodiment of a word voice recognition device according to the present invention.

FIG. 2 is a diagram for explaining the principle of vowel cutout in the embodiment shown in FIG.

FIG. 3 is a list of station names grouped according to predetermined vowels for the last syllable in the embodiment shown in FIG.

FIG. 4 is a list of station names grouped according to predetermined vowels for the frontmost syllable and the rearmost syllable in the embodiment shown in FIG.

[Explanation of symbols]

 1 voice output unit 2 A / D conversion unit 3 backup memory 4 vowel cutout unit 5 vowel storage unit 6 word storage unit 7 first stage voice recognition unit 8 second stage voice recognition unit 9 signal processing unit

Claims (4)

[Claims] 1. A vowel data storage means for storing data about a predetermined vowel, and a plurality of word data according to a type of vowel contained in a predetermined syllable among syllables forming each word. Vowel group storage means for classifying and storing into vowel groups, vowel extraction means for extracting vowels including the predetermined syllable of the input word speech as basic vowels, and basic vowel data extracted by the vowel extraction means , Comparing the data stored in the vowel data storage means,
First recognizing means for recognizing a vowel having the highest degree of similarity as the basic vowel, data of the input word voice, and data of words classified as the basic vowel recognized by the first recognizing means. A second speech recognition device for comparing and outputting a word having the highest degree of similarity as a recognition word. 2. The word voice recognition device according to claim 1, wherein the predetermined syllable is the last syllable of the syllables forming the word. 3. The word voice recognition device according to claim 1, wherein the predetermined syllable is the frontmost syllable of the syllables forming the word. 4. The word voice recognition device according to claim 1, wherein the predetermined syllables are two syllables of a front part and a last part of syllables forming the word.