JPH0981185A - Continuous voice recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a continuous voice recognition device which performs continuous voice recognition of natural uttering with a smaller computing cost than the one using a conventional method. SOLUTION: A word collating section 4 detects a word-hypothesis of an uttered voice sentence based on the feature parameters of the voice signals of the inputted uttered voice sentence using a one pass.viterbi decoding method, for example, computes the liklihood and outputs them. A word narrowing-down section 6 performs word-hypothesis narrowing-down so that the word-hypotheses of the same word outputted from the section 4 through a buffer memory 5 having the same completion time and a different starting time are represented by a single word-hypothesis having a highest liklihood among the entire liklihood computed from the starting time of uttering to the completion time of the work for every leading phoneme environment of the word.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a continuous voice recognition device for continuously recognizing a voice based on an input voice signal of a voiced voice sentence.

[0002]

2. Description of the Related Art Conventionally, the present applicant has been developing a continuous speech recognition system (hereinafter, referred to as a first conventional example) for the purpose of spontaneous speech recognition (for example, conventional document 1). "Nagai, Takami, Sagayama,
"The SSS-LR Continuous Sp
ech Recognition System: I
ntegrating SSS-Derivrd Al
lopohne Models and a Phon
eme-Context-DependentLR P
“Arser”, Proc. of ICSLP92, p.
p. 1511-1514, 1992 "and conventional literature 2
"Shimizu, Monzen, Singer, Ma
Tsunaaga, "Time-Synchronous
Continuous Speech Recogn
izer Driven bya Context-F
ree Grammar ”, Proc. of ICAS
SP95, pp. 584-587, 1995 ". ). In the first conventional example, a history of words from the start of utterance is used based on an input voice signal of a generated voice sentence by using a phoneme hidden Markov model (hereinafter, the hidden Markov model is referred to as HMM) and a word dictionary. Also, voice recognition is performed while managing the grammar state.

On the other hand, a speech recognition method using a word graph (hereinafter referred to as a second conventional example) is disclosed in a conventional document 3 "Ne.
y, Aubert, “A Word Graph Al
gorithm for Large Vocabul
ary, Continuous Speech Re
Cognition ”, Proc. of ICSLP9
4, pp. 1355-1358, 1994 "and the conventional document 4" Woodland, Leggetter, Od. "
ell, Valtchev, Young, “The 1
994 HTK Large Vocabulary
Speech Recognition System
m ", Proc. of ICASSP95, pp. 7
3-76, 1995 ".

The main idea of the word graph of the second conventional example is to process word hypothesis candidates in a region of a speech signal where ambiguity in speech recognition is relatively high. The advantage of this is that pure speech recognition is separate from the application of language models and that complex language models can be applied to the known steps following the word currently being recognized. The number of word hypothesis candidates must change according to the level of ambiguity in speech recognition. The difficulties in efficiently building a good word graph are as follows. The start time of a word is
Generally, it depends on the preceding word. In the first approximation, this dependency is limited to the immediately preceding word to obtain a so-called word pair approximation method as shown below. That is, given a word pair and its end time, the word boundary between two words is independent of another preceding word. This word pair approximation method has been introduced in order to efficiently calculate multiple sentences or n best sentences by nature. This word graph is expected to be more efficient than the approach of obtaining n bests (hereinafter referred to as the n-best method). In this method using a word graph, it is necessary to generate a plurality of word hypotheses only locally, whereas in the n-best method, each local word hypothesis candidate is a list of n best sentences. You need the whole sentence to add to.

[0005]

However, in the first conventional example, since it is necessary to manage the history of words and the grammatical state from the start of utterance, insertion of interjections, stagnation, and rewording are frequently performed. There is a problem that the calculation cost required for merging or dividing word hypotheses is extremely large when used for recognition of natural speech that occurs. That is, the processing amount required for voice recognition becomes large and a storage device having a comparatively large storage capacity is required, while the processing amount becomes large, resulting in a long processing time.

In the second conventional word pair approximation method, each preceding word is represented by one hypothesis.
The approximation effect is still relatively small. Therefore, the same problems as those of the first conventional example occur.

An object of the present invention is to solve the above problems and to provide a continuous speech recognition apparatus capable of performing continuous speech recognition of natural speech at a smaller calculation cost as compared with the conventional example.

[0008]

A continuous speech recognition apparatus according to the present invention detects a word hypothesis of an uttered voice sentence based on an input voice signal of the uttered voice sentence and calculates a likelihood to continuously output the word hypothesis. In a continuous speech recognition device having a speech recognition means for recognizing speech, the speech recognition means, for the word hypothesis of the same word with the same end time but different start time, for each head phoneme environment of the word, It is characterized in that the word hypotheses are narrowed down so as to be represented by one word hypothesis having the highest likelihood of the calculated total likelihood from the utterance start time to the end time of the word.

[0009]

DETAILED DESCRIPTION OF THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a block diagram of a continuous speech recognition apparatus according to an embodiment of the present invention. The continuous speech recognition apparatus of the present embodiment uses a known one-pass Viterbi decoding method to detect the word hypothesis of the uttered voice sentence based on the characteristic parameters of the voice signal of the input uttered voice sentence. In a continuous speech recognition apparatus having a word matching unit 4 for calculating and outputting a degree, a word hypothesis of the same word output from the word matching unit 4 via the buffer memory 5 and having the same end time but different start time is used. On the other hand, for each head phoneme environment of the word, the word hypothesis is represented by one word hypothesis having the highest likelihood of the total likelihood calculated from the utterance start time to the end time of the word. It is characterized in that a word hypothesis narrowing unit 6 for narrowing down is provided.

In FIG. 1, connected to the word matching unit 4,
For example, the phoneme HMM1 stored in the hard disk memory
1 is represented including each state, and each state has the following information. (A) State number (b) Acceptable context class (c) List of preceding states and succeeding states (d) Parameter of output probability density distribution (e) Probability of self transition and transition probability to subsequent states The phoneme HMM used in the example needs to specify which speaker each distribution is derived from, and thus is created by converting a predetermined speaker mixed HMM. Here, the output probability density function is a Gaussian mixture mixture having a 34-dimensional diagonal covariance matrix.

The word dictionary 12 connected to the word matching unit 4 and stored in, for example, a hard disk is a phoneme HMM.
For each of the 11 words, a symbol string indicating the symbolic reading is stored.

In FIG. 1, the vocalized voice of the speaker is input to the microphone 1 and converted into a voice signal, and then input to the feature extraction unit 2. The feature extraction unit 2 performs, for example, LPC analysis after A / D conversion of the input voice signal, and a 34-dimensional feature parameter including logarithmic power, 16th-order cepstrum coefficient, Δ logarithmic power, and 16th-order Δ cepstrum coefficient. To extract. The time series of the extracted characteristic parameters is input to the word matching unit 4 via the buffer memory 3.

The word collation unit 4 uses the one-pass Viterbi decoding method to generate a word hypothesis using the phoneme HMM 11 and the word dictionary 12 based on the data of the characteristic parameter input via the buffer memory 3. Is detected and the likelihood is calculated and output. Here, the word matching unit 4 determines that each HMM at each time
The likelihood within a word and the likelihood from the start of utterance are calculated for each state. The likelihood is individually provided for each word identification number, word start time, and difference between preceding words. Further, in order to reduce the amount of calculation processing, the grid hypothesis having a low likelihood of the total likelihood calculated based on the phoneme HMM 11 and the word dictionary 12 is reduced. The word matching unit 4 outputs the resulting word hypothesis and likelihood information to the word hypothesis narrowing unit 6 via the buffer memory 5 together with time information (specifically, for example, frame number) from the utterance start time. .

The word hypothesis narrowing unit 6 is based on the word hypothesis output from the word collating unit 4 via the buffer memory 5 for the word hypotheses of the same word having the same end time but different start times. For each head phoneme environment of a word, the word hypotheses were narrowed down to be represented by one word hypothesis having the highest likelihood of the total likelihood calculated from the vocalization start time to the end time of the word. After that, of the word strings of all the word hypotheses after narrowing down, the word string of the hypothesis having the maximum total likelihood is output as the recognition result. In the present embodiment, preferably, the first phoneme environment of the word to be processed is the three phonemes including the final phoneme of the word hypothesis preceding the word and the first two phonemes of the word hypothesis of the word. Say a line.

[0015] For example, as shown in FIG. 2, the (i-1) th word W _i-1 of the following phoneme string a _1, a _2, ..., come i th word W _i consisting a _n Sometimes, six hypotheses Wa, Wb, Wc, Wd, We, Wf are used as word hypotheses for the word W _i-1.
Exists. Here, the former three word hypotheses Wa, W
It is assumed that the final phonemes of b and Wc are / x /, and the final phonemes of the latter three word hypotheses Wd, We and Wf are / y /. Of the hypotheses in which the end time t _e is equal to the head phoneme environment (in FIG. 2, the top phoneme environment is “x / a ₁ / a ₂ ”, the three word hypotheses from the top), the hypothesis with the highest total likelihood (for example, FIG. Two
Are deleted except for the top hypothesis). Since the fourth phoneme from the top has a different first phoneme environment, that is, the last phoneme of the preceding word hypothesis is y instead of x,
Do not delete the fourth hypothesis from the top. That is, only one hypothesis remains for each final phoneme of the preceding word hypothesis. In the example of FIG. 2, one hypothesis is left for the final phoneme / x /, and one hypothesis is left for the final phoneme / y /.

In the above embodiment, the leading phoneme environment of the word is defined as three phoneme sequences including the final phoneme of the word hypothesis preceding the word and the first two phonemes of the word hypothesis of the word. Although defined, the present invention is not limited to this, and the phoneme string of the preceding word hypothesis including the final phoneme of the preceding word hypothesis and at least one phoneme of the preceding word hypothesis continuous with the final phoneme, and the word May be a phoneme sequence including a phoneme sequence including the first phoneme of the word hypothesis.

[0017]

EXAMPLES The present inventor conducted a word graph generation experiment using a spontaneous speech database in order to confirm the effectiveness of the continuous speech recognition apparatus of FIG. Spoken language database owned by the applicant with “travel planning” as a task (for example, refer to the conventional document 5 “Morimoto et.
al. , "A Speech and Langua
ge Database for Speech Tr
translationResearch ”, Proc.o
f ICSLP94, pp. 1791-1794, 19
1994 ". )) "Hotel reservation" dialogue (voices from 5 speakers of the applicant side: 5 dialogues, 56 vocalizations, 687 words). Acoustic analysis uses a sampling frequency of 12 kHz,
Analysis is performed with the specifications of a frame interval of 5 msec and a Hamming window of 20 msec, and 1 to 16th-order LPCs are used as characteristic parameters.
Cepstrum, 1st to 16th order ΔLPC cepstrum, logarithmic power, and Δlogarithmic power were used. The acoustic model (Hidden Markov network: 401 states, 5 mixed)
Utterances of nine speakers who did not appear in the test data of the above database (128)
Utterance) was used to adapt to the speaking style. Also, the language model is "travel planning" including "hotel reservation".
Students learned by using the general language (18,315 utterances, 229,159 words). The word perplexity was 55.9. The word dictionary (1,113 words) includes all vocabularies of the evaluation data, and there is no unknown word (also referred to as unregistered word) that is not registered in advance.

Next, the effect of narrowing down word hypotheses having different start times will be described below. FIG. 3 shows a comparison of the distribution of the number of preceding words of each word hypothesis when narrowing down (this embodiment) and when not narrowing down. By narrowing down, the average number of preceding words is 3.59.
Was reduced from 1.70 to 1.70. Further, the average number of preceding words was calculated to be 1.36 when the difference in the start time was ignored when the narrowing down was not performed. From this result, the method of the present invention in which one hypothesis is represented for each head phoneme environment of a word is 1 for each preceding word with a small calculation amount.
It is considered that an effect quite similar to the word pair approximation method of the second conventional example represented by one hypothesis can be obtained.

As described above, according to this embodiment, with respect to the word hypothesis of the same word having the same end time but different start time, for each head phoneme environment of the word, from the utterance start time to the word start of the word. The word hypotheses are narrowed down so as to be represented by one word hypothesis having the highest likelihood among the calculated total likelihoods up to the end time. That is, as compared with the second conventional word pair approximation method in which each preceding word is represented by one word hypothesis, those having the same leading phoneme of the first phoneme of the word (that is, the last phoneme of the preceding word) are collectively collected. You can reduce the number of word hypotheses to handle,
The approximation effect is large. In particular, the reduction effect is great when the number of vocabularies is increased. Therefore, even when the continuous speech recognition device is used for the recognition of natural utterances in which interjections are inserted, stagnation, and rewording frequently occur, the calculation cost required for merging or dividing word hypotheses is the same as in the conventional example. It becomes small compared to. That is, the amount of processing required for speech recognition is reduced, and therefore the working memory (not shown) of the word matching unit 4, the buffer memory 5 and the working memory of the word hypothesis narrowing unit 6 (not shown). While the storage capacity required in the storage device for voice recognition such as) is small, the processing amount is small, so that the processing time for voice recognition can be shortened.

[0020]

As described in detail above, according to the present invention, the word hypothesis of the uttered voice sentence is detected based on the input voice signal of the uttered voice sentence and the likelihood is calculated to continuously calculate the likelihood. In a continuous speech recognition device equipped with a speech recognition means, the speech recognition means starts utterance for each head phoneme environment of the word with respect to a word hypothesis of the same word having the same end time but different start times. Narrowing is performed so that one word hypothesis having the highest likelihood of the calculated total likelihood from the time to the end time of the word is represented. That is, as compared with the second conventional word pair approximation method in which each preceding word is represented by one word hypothesis, those having the same leading phoneme of the first phoneme of the word (that is, the last phoneme of the preceding word) are collectively collected. The number of word hypotheses can be reduced and the approximation effect is large. In particular, the reduction effect is great when the number of vocabularies is increased. Therefore, even when the continuous speech recognition device is used for the recognition of natural utterances in which interjections are inserted, stagnation, and rewording frequently occur, the calculation cost required for merging or dividing word hypotheses is the same as in the conventional example. It becomes small compared to. In other words, the amount of processing required for voice recognition is reduced, and therefore the storage capacity required in the storage device for voice recognition is reduced, while the amount of processing is reduced, which reduces the processing time for voice recognition. can do.

[Brief description of drawings]

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

2 is a timing chart showing a process of a word hypothesis narrowing unit 6 in the continuous speech recognition apparatus of FIG.

3 is a graph of the number of nodes with respect to the number of preceding words showing the effect of narrowing the word hypothesis in the transition between words in the experimental result of the continuous speech recognition apparatus of FIG.

[Explanation of symbols]

 1 ... Microphone, 2 ... Feature extraction part, 3, 5 ... Buffer memory, 4 ... Word collation part, 6 ... Word hypothesis narrowing part, 11 ... Phoneme HMM, 12 ... Word dictionary.

Front page continued (72) Inventor Shoichi Matsunaga Seika-cho, Soraku-gun, Kyoto Pref. 5 Inhiratani, Mihiratani No.5, A.T.R. Co., Ltd. Speech Translation Communication Research Laboratories (72) Inventor Yoshinori Kosaka, Seiraku-cho, Soraku-gun, Kyoto Prefecture Daiji Intani, Shoji, Hiratani, No.5, ATR Co., Ltd.

Claims (1)

[Claims] 1. Continuous speech recognition provided with a speech recognition means for continuously recognizing speech by detecting a word hypothesis of the uttered speech sentence based on an input speech signal of the uttered speech sentence and calculating a likelihood. In the device, the speech recognition means calculates, for each of the head phoneme environments of the word, from the utterance start time to the end time of the word for the word hypothesis of the same word having the same end time but different start times. A continuous speech recognition device characterized in that a word hypothesis is narrowed down so that it is represented by one word hypothesis having the highest likelihood of the total likelihood.