JPH09292899A - Voice recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the recognition utilizing continuation time of syllable by correcting a vice, which is as candidate of recognition result, on the basis of the mean syllable length and the continuation time length of each syllable of the voice to be recognized. SOLUTION: An input voice is also input to a syllable boundary detecting unit 104, and the syllable detecting unit 104 obtains a position of candidate of a syllable boundary on the basis of the power information or the like, and outputs it to a mean syllable length estimating unit 105. The mean syllable length estimating unit 105 utilizes a characteristic that the continuation time length of each syllable is nearly constant so as to estimate the mean syllable length of the input voice on the basis of the information on each syllable boundary candidate. A candidate selecting unit 106 checks whether the continuation time length of each syllable exists between the minimum continuation time and the maximum continuation time length, which are decided on the basis of the mean syllable length obtained by the mean syllable length estimating unit 105, or not for each candidate, which are output from a voice recognizing unit 101, and in the case where any one sound, of which continuation time is out of the range, exists, this candidate is discarded, and in the case where a candidate, of which all syllables having continuation time in the range, is found, this candidate is recognized.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition apparatus, and more particularly to a speech recognition apparatus using a duration time of an acoustic unit such as a syllable.

[0002]

2. Description of the Related Art Conventionally, DP (dynamic programming) matching and hidden Markov model (H) are used.
A voice recognition device using "MM") has been put to practical use. Here, the DP matching will be briefly described. For example, when calculating the spectral distance over the entire voice section of the standard pattern and the input voice, the time axis normal that expands and contracts the time axis so that the same phonemes of both are corresponded. It is for performing conversion. Further, the HMM is a method of forming a transition diagram (Markov model) consisting of a small number of states for each word or each phoneme, and recognizing by recognizing which model the input speech is most likely generated. , It is called "hidden" because what is observed at this time is the spectrum sequence generated by the transition and the state itself is not observed. For each model, the probability of occurrence of spectral parameters in each state and the transition probability between states are estimated using learning samples. At the time of recognition, the input speech is applied to each model and the probability of generating the input speech is highest. Select a model as the recognition result.

In a speech recognition apparatus using this DP matching or a hidden Markov model, when matching an input pattern and a standard pattern with a subword such as a syllable as a recognition unit, information on the duration of each recognition unit is used. There is a method for improving recognition performance. For example, a method is generally known in which the maximum value and the minimum value of the duration time for each phoneme are set in advance and the matching section is restricted by this.

A conventional technique of this type will be described below with reference to FIG.

Referring to FIG. 7, the input voice is input to the voice recognition unit 701 and converted into a time series of feature vectors. The speech recognition unit 701, for each word in the word dictionary 702, a word standard pattern in which phoneme standard patterns 703 are connected based on reading information written in the dictionary 702,
The matching with the input voice converted into the time series of the feature vector is performed using DP matching. At that time, the maximum and minimum values of the duration of each phoneme are read from the phoneme duration information 704, and matching in which the duration of the phoneme is out of the range is prohibited. As a result of matching, the word with the smallest distance is output as the recognition result.

The maximum value and the minimum value of the phoneme duration are obtained in advance by using the learning data with phoneme labels. This eliminates inappropriate matching due to extreme expansion and contraction of the time axis. Further, for example, it becomes easier to recognize words whose difference is mainly in the phoneme duration, such as “grandmother” and “grandmother”, and the recognition performance is improved.

[0007]

The above-mentioned conventional method is effective when uttered at a certain fixed speed, but the maximum and minimum values of the duration of the phoneme or syllable depend on the generation speed. change. That is, in the above-mentioned conventional method, the maximum and minimum values of the phoneme duration are fixed and used, so that they are weak against the fluctuation of the speaking rate (that is, the recognition performance is easily affected by the fluctuation). ) Has a problem.

Therefore, the present invention has been made in view of the above circumstances, and an object thereof is to provide a voice recognition device which has a high recognition performance by utilizing a duration time and is strong against fluctuations in utterance speed. To do.

[0009]

In order to achieve the above object, a speech recognition apparatus according to the present invention obtains an average syllable length from input speech to be recognized, and continues the average syllable length and each syllable of a recognition result candidate. It is characterized by selecting from the recognition result candidates or correcting the recognition result based on the time length.

A voice recognition device according to the present invention recognizes an input voice, adds a plurality of recognition result candidates with information on the duration of each syllable, and outputs the voice recognition means, and a syllable from the input voice. Syllable boundary candidate detecting means for finding boundary candidates,
An average syllable length estimating means for obtaining an average syllable length from the syllable boundary candidate, and a candidate selecting means for selecting a recognition result from the plurality of recognition result candidates based on the recognition result candidate and the average syllable length.

Further, the voice recognition device according to the present invention recognizes an input voice, outputs a plurality of recognition result candidates by adding segmentation for each syllable or information on the duration of each syllable, and outputting the result. And an average syllable length estimating means for obtaining an average syllable length for each candidate of the recognition result, and a candidate selecting means for selecting a recognition result from the plurality of recognition result candidates based on the recognition result candidate and the average syllable length. ,
including.

According to the present invention, since the candidate of the recognition result is selected or corrected using the average syllable length, the candidate which is improperly matched due to the extreme expansion and contraction of the time axis at the time of recognition is rejected or corrected and recognized. Since the performance is improved and the average syllable length is obtained from the input voice itself, it is less susceptible to the fluctuation of the speaking speed.

[0013]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

A first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention.

The input voice is input to the voice recognition unit 101,
It is converted into a time series of feature vectors. Voice recognition unit 101
Is a word dictionary 10 for each word in the word dictionary 102.
Syllable standard pattern 1 based on the reading information written in 2.
A standard word pattern in which 03 is concatenated is compared with an input voice converted into a time series of feature vectors to obtain a matching score and segmentation for each syllable. For matching, for example, DP matching can be used. Alternatively, the syllable standard pattern can be represented by a hidden Markov model, and the Viterbi algorithm can be used for matching.

Then, a plurality of recognition result candidates are selected based on the score of the matching result and output to the candidate selecting unit 106 together with the matching score and the information about the duration of each syllable. The duration of each syllable can be easily calculated from the segmentation information for each syllable.

For example, when uttering "grandmother",
An example of the recognition result candidates is shown in FIG. The voice recognition unit 1
In 01, continuous speech recognition of a sentence or a phrase can also be performed using a syntax rule together with a word dictionary.

On the other hand, referring again to FIG. 1, the input voice is also input to the syllable boundary detecting unit 104, and the syllable boundary detecting unit 10
4 calculates a candidate for the position of the syllable boundary using information such as power and outputs it to the average syllable length estimation unit 105.

An example of syllable boundary candidates is shown in FIG. The syllable boundary candidates may include erroneous syllable boundaries or may not include correct syllable boundaries. However, in the average syllable length estimation unit 105 (see FIG. 1), the duration length of each syllable is By utilizing the fact that it is almost constant, the average syllable length of the input speech is estimated based on the information of the syllable boundary candidates.

When a section between adjacent syllable boundary candidates is regarded as a speech section candidate, a syllable section candidate extremely longer than the time length of another syllable section candidate actually includes a plurality of syllables. It is considered that the extremely short syllable section candidates are one syllable together with the adjacent syllable section candidates.

For example, in the case of the example of the syllable boundary candidate shown in FIG. 3, the interval between the second syllable boundary candidate and the third syllable boundary candidate (the interval of 240 to 540 msec) is actually 2 It is considered to include syllables. With this in mind,
In the example of FIG. 3, the average syllable length is estimated to be about 150 msec. Specifically, the average syllable length is estimated, for example, by the method described below.

Assuming the syllable length, the number of syllables, and the starting point of the first syllable, all syllable boundaries are determined. Therefore, this is compared with the syllable boundary candidates output by the speech boundary detection unit 104, and the degree of similarity is obtained. .

The degree of similarity is calculated by presetting a penalty for insertion, dropout, and positional deviation of syllable boundary candidates.

The syllable length, the number of syllables, and the starting point of the first syllable are changed to find the one with the highest degree of similarity, and the syllable length at that time is taken as the average syllable length. To estimate the average syllable length, use the average syllable length estimated for previous utterances, and add the difference between this average syllable length and the average syllable length of the utterance as a penalty when determining the similarity. You can also

In the candidate selection unit 106, the voice recognition unit 101
For each candidate output by, whether the duration of each syllable is between the minimum duration and the maximum duration determined based on the average syllable length obtained by the average syllable length estimation unit 105 in the order of the matching score. To find out.

If there is even one syllable whose duration is not within this range, the candidate is rejected.

When a candidate whose duration time is within the above range is found for all syllables, this is output as a recognition result, and the selection is completed.

The minimum duration and the maximum duration are set, for example, so that the difference or ratio with the average syllable length is a predetermined value. This difference or ratio may use a common value for each syllable, or more precisely, a different value for each syllable type.

For example, for a given input voice, a plurality of recognition result candidates as shown in FIG. 2 are obtained, and the average syllable length is 1
It is assumed to be 50 msec. With reference to FIG. 2, a syllable and length information of the syllable are stored for each recognition result candidate. The minimum and maximum durations are
If the difference from the average syllable length is set within 30 msec, the minimum duration is 120 msec and the maximum duration is 180 msec.

Of the plurality of recognition result candidates shown in FIG. 2, the first candidate is rejected because the second syllable (310 msec) exceeds the maximum duration (180 msec),
The second candidate is selected and output as a recognition result.

Further, the candidate selecting section 106 can be configured as follows. That is, for each candidate output by the voice recognition unit 101, a duration score is obtained for each syllable based on the duration and the average syllable length obtained by the average syllable length estimation unit 105.

These continuation time scores and the voice recognition unit 1
The total score of the candidates is obtained from the matching score obtained in 01,
The candidate with the highest total score is output as the recognition result.
As a method of obtaining the duration time score, for example, a distribution regarding the duration time of the syllable can be determined based on the average syllable length, and the likelihood can be obtained as the likelihood for the duration time of each syllable based on the distribution. The distribution of the duration can be determined, for example, by assuming that the variance is a Gaussian distribution in which each syllable type is determined in advance, and taking the average value as the average syllable length.

Next, a second embodiment of the present invention will be described. FIG. 4 is a block diagram showing the configuration of the second embodiment of the present invention.

Referring to FIG. 4, the input voice is input to the voice recognition unit 401, and the voice recognition unit 401 performs matching as in the case of the first embodiment, and a plurality of recognition result candidates are obtained. The matching score and the segmentation information for each syllable are sequentially output to the candidate selecting unit 405 and the average syllable length estimating unit 404 in the order of the matching score.

The average syllable length estimation unit 404 treats, for each candidate of the recognition result, the boundary of the syllable by the segmentation in the same manner as the syllable boundary candidate in the first embodiment, and thus the average syllable of the input voice. Estimate the length.

Since the average syllable length can be estimated only from the relative positional relationship of syllable boundaries, the duration of each syllable instead of segmentation for each syllable is added as information to the recognition result candidates. Information may be used.

In the candidate selecting section 405, the duration of each syllable obtained from the segmentation for each recognition result candidate is determined based on the average syllable length obtained by the average syllable length estimating section 404. To see if it is in between.

If there is even one syllable whose duration is not within this range, the candidate is rejected. When a candidate whose duration is within the range is found for all syllables, the candidate is output as a recognition result, and the selection ends.

The operation of the candidate selecting section 405 is the same as that of the first embodiment except that the average syllable length differs for each syllable candidate.

Further, the average syllable length obtained for the first candidate can be used for other candidates.

Further, as described in the first embodiment, it is also possible to obtain a continuation time score for each recognition result candidate and select it by the total score.

Next, a third embodiment of the present invention will be described. FIG. 5 is a block diagram showing the configuration of the third exemplary embodiment of the present invention.

Referring to FIG. 5, the input voice is input to the voice recognition unit 501 and converted into a time series of feature vectors. The voice recognition unit 501 recognizes the input voice as an arbitrary syllable string using the syllable standard pattern 503 and the syllable string connection rule 502.

Here, the syllable connection rule is, for example, ""
Represents a general connection rule of Japanese syllables such as "is not continuous and does not appear in the beginning of words", "vowels do not come after""and is preferably expressed by a finite state network. can do.

It is also possible to give a probability to each transition of the network.

The speech recognition unit 501 performs recognition by the continuous speech recognition algorithm under finite state network control, and uses the syllabic sequence with the best score as the temporary recognition result, with the segmentation information for each syllable as the recognition result correction unit 506. It outputs to the average syllable length estimation unit 505.

The average syllable length estimation unit 505 estimates the average syllable length of the input voice based on the segmentation information of the temporary recognition result, as in the second embodiment.

The recognition result correction unit 506 compares the duration time of each syllable obtained by the segmentation with the average syllable length obtained by the average syllable length estimation unit for the temporary recognition result. When the ratio of the duration to the average syllable length is not within a predetermined range, the syllable string correction rule 504 is referred to replace the syllable string containing the syllable or an adjacent syllable with another syllable or syllable string. As a result, the temporary recognition result is corrected. The character string obtained as a result of the correction is output as the recognition result.

FIG. 6 shows an example of a syllable string correction rule according to the third embodiment of the present invention.

Referring to FIG. 6, the syllable string correction rule includes a division rule and a merging rule. 2 duration is the average syllable length
In the case of about twice or more, according to the division rule, the syllable is replaced with the corrected syllable string.

For syllables whose duration is less than about half of the average syllable length and which becomes almost the average syllable length together with the durations of adjacent syllables, according to the merge rule,
The syllables adjacent to the syllable are collectively replaced with one syllable.

FIG. 6 shows an example of a very simple syllable sequence correction rule. However, in practice, the application of the rule may be restricted or given a priority condition depending on the duration and the syllable environment before and after, or a more complicated correction may be made. It is also possible to give rules.

In the third embodiment of the present invention, instead of obtaining the average syllable length from the tentative recognition result, as in the first embodiment, a syllable boundary detecting unit is provided and the syllable boundary candidates to be output are provided. It is also possible to ask from.

In the above-described embodiment of the present invention,
Although syllables are used as the speech recognition unit and the unit for obtaining the duration time, it is not limited to syllables, and any acoustic unit such as a phoneme or a semi-syllable can be used.

Further, in the present invention, the voice recognition unit and the unit for obtaining the duration are not necessarily the same. For example, a phoneme is used as the voice recognition unit and a syllable is used as the unit for obtaining the duration. You can also Further, when obtaining the average syllable length, instead of obtaining from the syllable boundary candidates, for example, it is possible to obtain from the candidate of the central part of the vowel of the syllable.

[0056]

As described above, according to the present invention,
The present invention realizes a voice recognition device capable of high-performance recognition using the duration of syllables and the like, regardless of fluctuations in vocalization rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of a recognition result candidate according to the first embodiment of the present invention.

FIG. 3 is a diagram showing an example of syllable boundary candidates in the first embodiment of the present invention.

FIG. 4 is a block diagram showing a second embodiment of the present invention.

FIG. 5 is a block diagram showing a third embodiment of the present invention.

FIG. 6 is a diagram showing an example of a voice correction rule according to a third embodiment of the present invention.

FIG. 7 is a block diagram of a conventional example.

[Explanation of symbols]

 101 voice recognition unit 102 word dictionary 103 syllable standard pattern 104 syllable boundary detection unit 105 average syllable length estimation unit 106 candidate selection unit

Claims (10)

[Claims] 1. A speech recognition apparatus, characterized in that a recognition result is selected from recognition result candidates for an input speech based on a duration of each syllable of the input speech and an average syllable length obtained from the input speech. . 2. A speech recognizing means for recognizing an input voice and outputting a plurality of recognition result candidates by adding information on a duration time of each syllable, and a syllable boundary candidate for obtaining a syllable boundary candidate from the input voice. A detection means; an average syllable length estimation means for obtaining an average syllable length from the syllable boundary candidates; and a candidate selection means for selecting a recognition result from the plurality of recognition result candidates based on the recognition result candidates and the average syllable length. A voice recognition device comprising: 3. A speech recognition means for recognizing an input speech, outputting a plurality of recognition result candidates by adding segmentation for each syllable or duration information of each syllable, and outputting the result, and each candidate of the recognition result. An average syllable length estimating means for obtaining an average syllable length; and a candidate selecting means for selecting a recognition result from the plurality of recognition result candidates based on the recognition result candidate and the average syllable length. Recognition device. 4. The average syllable length obtained for the highest recognition result candidate is used for all recognition result candidates instead of obtaining the average syllable length for each candidate recognition result by the average syllable length estimation means. 4. The method according to claim 3, wherein
The speech recognition device according to the above. 5. The average syllable length estimating means holds information on the average syllable length obtained for past input speech, and together with this, obtains the average syllable length of the input speech. Item 5. The voice recognition device according to any one of items 2 to 4. 6. The candidate selecting means obtains at least one of a maximum duration and a minimum duration of syllables from the average syllable common to all syllables or for each syllable type, and the maximum duration is It is determined whether or not to reject each candidate of the recognition result candidates based on at least one value of the minimum duration, and the highest candidate among the candidates that are not rejected is set as the recognition result. The voice recognition device according to any one of 2 to 5. 7. The speech recognition means adds the information of the matching score to the recognition result candidate and outputs it, and the candidate selecting means, for each candidate of the recognition result candidate, the duration time of each syllable thereof. And a duration time score based on an average syllable length, and a recognition result is selected from the recognition result candidates based on the matching score and the duration time score of each candidate of the recognition result candidates. The voice recognition device according to any one of 2 to 5. 8. A voice recognition device, which corrects a recognition result for an input voice based on a duration time of each syllable, an average syllable length obtained from the input voice, and a syllable string correction rule. 9. A voice recognizing means for recognizing an input voice and outputting a tentative recognition result by adding information on a duration time of each syllable, and a syllable boundary candidate detecting means for obtaining a syllable boundary candidate from the input voice. An average syllable length estimating means for obtaining an average syllable length from the syllable boundary candidate; and a recognition for obtaining a recognition result by correcting the temporary recognition result based on the temporary recognition result, the average syllable length and a syllable string correction rule. A voice recognition device comprising: result correction means; 10. A voice recognition means for recognizing an input voice and outputting a provisional recognition result by adding segmentation for each syllable or information of duration of each syllable, and obtaining an average syllable length from the provisional recognition result. An average syllable length estimating means, and a recognition result correcting means for correcting the temporary recognition result based on the temporary recognition result, the average syllable length, and a syllable string correction rule to obtain a recognition result. Speech recognizer.