JP2007127738A - Voice recognition device and program therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the following problem; a conventional voice recognition device sometimes outputs a misrecognition result, and in such a case, users' trust in the voice recognition device cannot be built. <P>SOLUTION: The output of the misrecognition result can be prevented by the voice recognition device comprising: a voice input part which receives an voice input; a voice recognition part which performs voice recognition processing to the received voice using stored acoustic data, and acquires the result information of the voice recognition processing as a result of the voice recognition processing; a pronunciation evaluating part which acquires a pronunciation evaluation result by performing pronunciation evaluation to the received voice using the acoustic data; a judging part which judges whether or not the pronunciation evaluation result has a predetermined relation to a predetermined threshold value; an output part which outputs the voice recognition result about the result information of the voice recognition processing acquired by the voice recognition part only when the judging part judges that the pronunciation evaluation result has the predetermined relation. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a speech recognition apparatus that performs speech recognition processing.

In a conventional speech recognition device, there are speech recognition devices using various speech recognition algorithms for improving recognition accuracy (see, for example, Patent Document 1, Patent Document 2, Patent Document 3, and Patent Document 4).
JP 2003-44079 A (first page, FIG. 1 etc.) Japanese Unexamined Patent Publication No. 2003-22091 (first page, FIG. 1 etc.) Japanese Patent Laid-Open No. 10-254350 (first page, FIG. 1 etc.) JP-A-7-199989 (first page, FIG. 1 etc.)

  However, in the conventional speech recognition apparatus, even if speech recognition processing is performed with an algorithm with high recognition accuracy, an erroneous recognition result may be output. In such a case, the user's trust in the speech recognition apparatus is increased. There was a problem that it could not be built.

  When the pronunciation of the input speech is rated and the pronunciation rating result is low, the possibility of outputting an incorrect recognition result can be greatly reduced by a voice recognition device that does not output the recognition result. As a result, the user's trust for the speech recognition apparatus can be ensured.

  Specifically, this speech recognition apparatus is the following speech recognition apparatus.

  The speech recognition apparatus according to the first aspect of the invention uses an acoustic data storage unit that stores acoustic data that is data relating to a target voice to be compared, a voice input unit that receives voice input, and the acoustic data. A voice recognition unit that performs voice recognition processing on the voice received by the voice input unit and acquires voice recognition processing result information that is a result of the voice recognition processing; and a voice that is received by the voice input unit. A pronunciation rating unit that performs pronunciation rating processing using the acoustic data and acquires a pronunciation rating result; a determination unit that determines whether the pronunciation rating result has a predetermined relationship with a predetermined threshold; The speech recognition apparatus includes an output unit that outputs a speech recognition result for the speech recognition processing result information acquired by the speech recognition unit only when it is determined that the determination unit has a predetermined relationship.

  With this configuration, it is possible to greatly reduce the possibility of outputting an erroneous recognition result.

  The speech recognition apparatus according to the second aspect of the present invention is the speech recognition processing result obtained by the speech recognition unit with respect to the speech received by the speech input unit. This is a speech recognition device that performs pronunciation rating processing using information and the acoustic data, and acquires a pronunciation rating result.

  With this configuration, it is possible to greatly reduce the possibility of outputting an erroneous recognition result.

  The speech recognition apparatus according to the third aspect of the present invention is the voice recognition device according to any one of the first and second aspects, wherein the pronunciation rating unit includes frame classification means for classifying the voice received by the voice reception unit into frames. Frame audio data obtaining means for obtaining one or more frame audio data, which is audio data for each of the divided frames, optimum state determining means for determining an optimum state of the one or more frame audio data, and the optimum state determining means A sounding interval probability value acquisition means for acquiring the probability value of the optimum state determined by each sounding interval, and one or more probability values for each of the one or more sounding intervals acquired by the sounding interval probability value acquisition means as parameters, A speech recognition apparatus comprising rating value calculation means for calculating a rating value of speech.

  With this configuration, it is possible to perform pronunciation evaluation with higher accuracy, and to further greatly reduce the possibility of outputting erroneous recognition results.

  Further, in the voice recognition device according to the fourth aspect of the present invention, with respect to any one of the first and second aspects, the pronunciation rating unit includes frame classification means for classifying the voice received by the voice reception unit into frames. An optimum state determining means for determining an optimum state of the one or more frame sound data, and one or more probability values in the entire phoneme state having the optimum state of each frame determined by the optimum state determining means, for each sounding section. A speech section frame phoneme probability value acquiring means, and a rating value calculation for calculating a speech rating value using one or more probability values for each of one or more sound sections acquired by the sound section frame phoneme probability value acquiring means as parameters. A speech recognition device comprising means.

  With this configuration, it is possible to perform pronunciation evaluation with higher accuracy, and to further greatly reduce the possibility of outputting erroneous recognition results.

  The speech recognition apparatus according to the fifth aspect of the invention relates to either of the third and fourth aspects of the invention, wherein the pronunciation rating unit includes silence data storage means for storing silence data that is data indicating silence. , Further comprising a silent section detecting means for detecting a silent section based on the input voice data and the silent data, wherein the optimum state determining means is the one or more excluding the frame voice data for the silent section. This is a speech recognition apparatus for determining the optimum state of the frame speech data.

  With this configuration, it is possible to perform more accurate pronunciation evaluation by detecting silence data, and as a result, the possibility of outputting an erroneous recognition result can be further greatly reduced.

  According to the speech recognition apparatus of the present invention, it is possible to prevent an erroneous recognition result from being output.

  Hereinafter, embodiments of a speech recognition apparatus and the like will be described with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.

(Embodiment 1)
The speech recognition apparatus according to the present embodiment performs a pronunciation rating process for determining whether the received voice is good or bad, and the voice recognition result is obtained only when the rating value as a result of the pronunciation rating process satisfies a predetermined condition. Is a voice recognition device that outputs. Further, the rating value in the pronunciation rating process performed by the speech recognition apparatus is calculated for each pronunciation interval.

  It should be noted that, for the pronunciation evaluation process of each frame performed by the speech recognition apparatus, the posterior probability of the optimum state for the frame of the input speech is calculated using dynamic programming, so that the posterior probability is calculated by DAP (Dynamic A Postiori Probability). ), And a device that performs a similarity calculation method based on DAP and a pronunciation rating process is called DAPS. Then, since the speech recognition apparatus calculates a rating value for each sound generation section, the posterior probability calculated by the speech recognition apparatus is referred to as t-DAP with respect to DAP.

  FIG. 1 is a block diagram of the speech recognition apparatus according to the present embodiment.

  The speech recognition apparatus includes a reception unit 101, a speech reception unit 102, an acoustic data storage unit 103, a recognition candidate data storage unit 104, a speech recognition unit 105, a pronunciation rating unit 106, a determination unit 107, and an output unit 108.

  The pronunciation rating unit 106 includes frame classification means 1061, frame sound data acquisition means 1062, optimum state determination means 1063, pronunciation interval probability value acquisition means 1064, and rating value calculation means 1065.

  The accepting unit 101 accepts a start instruction that is an instruction to start processing of the speech recognition apparatus. Input means such as a start instruction may be anything such as a numeric keypad, a keyboard, a mouse, or a menu screen. The receiving unit 101 can be realized by a device driver for input means such as a numeric keypad or a keyboard, control software for a menu screen, and the like.

  The voice receiving unit 102 receives voice input. Audio is input from a microphone or the like. The input voice is information for voice recognition. The voice reception unit 102 can be realized by a microphone and driver software.

  The acoustic data storage unit 103 stores acoustic data that is data related to the target speech to be compared. The acoustic data is preferably, for example, data based on an HMM in which hidden Markov models for each phoneme are connected. The acoustic data is preferably data based on an HMM in which HMMs corresponding to phonemes constituting the input voice are linked according to the input order. However, the acoustic data does not necessarily need to be data based on the HMM obtained by connecting the HMMs for each phoneme. The acoustic data may be a simple set of all phoneme HMMs. The acoustic data does not necessarily need to be data based on the HMM. The acoustic data may be data based on other models such as a single Gaussian distribution model, a probability model (GMM: Gaussian mixture model), and a statistical model. The data based on the HMM has, for example, a state identifier and transition probability information for each frame. Further, the data based on the HMM may be, for example, a model learned (estimated) from two or more data uttered by a person speaking with a plurality of speech recognition target languages as a native language. The acoustic data storage unit 103 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium.

  The recognition candidate data storage unit 104 stores one or more recognition candidate data, which is data indicating voice recognition candidates. The recognition candidate data is, for example, a phoneme transcription of a word. Here, the phoneme transcription is information that represents a phoneme in a character string. The phoneme transcription includes, for example, “inu”, “neko”, “ume”, “hitsuji”, and the like. The recognition candidate data storage unit 104 may store information that represents one phoneme (for example, “u”) in a character string instead of a word. The recognition candidate data may be an identifier for identifying a speech recognition candidate, for example. When the recognition candidate data is an identifier for identifying a speech recognition candidate, the data indicating the speech recognition candidate itself is included in, for example, an external device or the present speech recognition device (not shown). The recognition candidate data storage unit 104 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium.

  The speech recognition unit 105 performs speech recognition processing on the speech received by the speech reception unit 102 using the acoustic data or the acoustic data and the recognition candidate data, and speech recognition processing result information that is a result of the speech recognition processing. To get. The speech recognition processing result information is, for example, recognition candidate data (for example, phoneme transcription) of the speech recognition processing result or an optimum state sequence of the speech recognition processing result. The data structure of the speech recognition processing result information does not matter. The voice recognition unit 105 normally performs voice recognition processing after the receiving unit 101 receives a start instruction. The voice recognition unit 105 may recognize specific words using the acoustic data and the recognition candidate data, or recognize unspecified words using the acoustic data without using the recognition candidate data. It may be processed. Moreover, the algorithm of a speech recognition process is not ask | required. Since various known techniques exist for the speech recognition process itself, the description thereof is omitted. The speech recognition unit 105 can usually be realized by an MPU, a memory, or the like. The processing procedure of the voice recognition unit 105 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The pronunciation rating unit 106 performs a pronunciation rating process on the voice received by the voice receiving unit 102 using the acoustic data in the acoustic data storage unit 103, and acquires a pronunciation rating result. In general, the pronunciation rating unit 106 performs a pronunciation rating process on the voice received by the voice receiving unit 102 using the voice recognition processing result information and the acoustic data acquired by the voice recognition unit 105, and obtains the pronunciation rating result. get. That is, normally, the pronunciation rating unit 106 performs the pronunciation rating process also using the voice recognition processing result information obtained as a result of the voice recognition in the voice recognition unit 105. Although details of the pronunciation rating process will be described later, various pronunciation rating processing algorithms are conceivable. The pronunciation rating unit 106 can usually be realized by an MPU, a memory, or the like. The processing procedure of the pronunciation rating unit 106 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The frame classification means 1061 constituting the pronunciation rating unit 106 classifies the voice received by the voice receiving unit 102 into frames. The processing of the frame sorting means 1061 is a processing according to a known technique.

  The frame audio data acquisition unit 1062 obtains one or more frame audio data that are audio data for each frame divided by the frame classification unit 1061.

Optimal state determination means 1063 determines an optimal state for at least one frame audio data of one or more frame audio data. The optimum state determination unit 1063 acquires, for example, an HMM corresponding to one or more phonemes constituting a speech such as a recognition target word from all phoneme HMMs in the acoustic data storage unit 103, and the acquired one or more HMMs Are concatenated in the order of phonemes (data related to the speech to be recognized and data based on the HMM concatenating hidden Markov models for each phoneme). Then, based on each feature vector o _t constituting constituting the relevant data, and the obtained feature vector series, to determine the optimal conditions for a given frame (optimum condition for the feature vector o _t). The algorithm for determining the optimum state is, for example, the Viterbi algorithm.

  The pronunciation interval probability value acquisition unit 1064 acquires the probability value of the optimum state determined by the optimum state determination unit 1063 for each pronunciation interval. Here, the pronunciation period is a period that constitutes a group of pronunciations such as phonemes, syllables, and words.

  The rating value calculation means 1065 calculates a speech rating value using one or more probability values for each of one or more pronunciation intervals acquired by the pronunciation interval probability value acquisition means 1064 as parameters. For example, the rating value calculation means 1065 calculates, for each sounding section, a time average value of one or more probability values of each sounding section acquired by the sounding section probability value acquisition means 1064 to obtain one or more time average values. The voice rating value is calculated using the one or more time average values as parameters.

  The frame segmentation means 1061, the frame sound data acquisition means 1062, the optimum state determination means 1063, the pronunciation interval probability value acquisition means 1064, and the rating value calculation means 1065 can be usually realized by an MPU, a memory, or the like. The processing procedure of the frame sorting means 1061 and the like is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The determination unit 107 determines whether the pronunciation evaluation result has a predetermined relationship with the threshold value. Here, the pronunciation rating result is usually a numerical value indicating the level of pronunciation rating. The pronunciation rating result is, for example, a numerical value from 0 to 1, or a numerical value in 10 steps from 1 to 10. The threshold is usually a numerical value stored in advance. In addition, the predetermined relationship means, for example, that the pronunciation evaluation result (for example, a numerical value from 0 to 1) is equal to or greater than a threshold value (for example, 0.50). The predetermined relationship means that, for example, a pronunciation evaluation result (for example, a numerical value from 0 to 1) is larger than a threshold value (for example, 0.50). The determination unit 107 can be usually realized by an MPU, a memory, or the like. The processing procedure of the determination unit 107 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The output unit 108 outputs the speech recognition result for the speech recognition processing result information acquired by the speech recognition unit 105 only when the determination unit 107 determines that there is a predetermined relationship. The voice recognition result may be the same information as the voice recognition process result information, or information generated from the voice recognition process result information. When the speech recognition processing result information is phoneme transcription (for example, “neko”), the speech recognition result is, for example, a character string (for example, “cat”). Further, when the determination unit 107 determines that the determination unit 107 does not have a predetermined relationship, the output unit 108 normally outputs information indicating that voice recognition has failed. However, in such a case, no output is required. Here, the output is a concept including display on a display, printing on a printer, sound output, transmission to an external device, accumulation in a recording medium, and the like. The output unit 108 may or may not include an output device such as a display or a speaker. The output unit 108 can be realized by output device driver software, or output device driver software and an output device.

  Next, the operation of the speech recognition apparatus will be described with reference to the flowcharts of FIGS.

  (Step S201) The receiving unit 101 determines whether a start instruction has been received. If a start instruction is accepted, the process proceeds to step S202. If a start instruction is not accepted, the process returns to step S201.

  (Step S202) The voice reception unit 102 determines whether a voice is received. If a voice is accepted, the process goes to step S203, and if no voice is accepted, the process returns to step S202. This voice is a voice to be recognized.

  (Step S203) The voice recognition unit 105 performs voice recognition processing on the voice received in step S202. Specifically, the speech recognition unit 105 performs speech recognition processing on the speech received in step S202 using the acoustic data and the recognition candidate data, and obtains speech recognition processing result information that is a result of the speech recognition processing. get.

  (Step S204) The pronunciation rating unit 106 performs a pronunciation rating process on the voice received by the voice receiving unit 102 using the acoustic data in the acoustic data storage unit 103, and acquires a pronunciation rating result. The pronunciation rating unit 106 normally performs a pronunciation rating process using acoustic data from the speech recognition process result information obtained in step S203, and acquires a pronunciation rating result. Details of the pronunciation rating process will be described with reference to the flowchart of FIG.

  (Step S205) The determination unit 107 determines whether or not the pronunciation rating result is in a predetermined relationship (for example, pronunciation rating result> threshold) with respect to the predetermined threshold. If there is a predetermined relationship, go to step S206, and if not, go to step S207.

  (Step S206) The output unit 108 outputs a speech recognition result (for example, a character string of the speech recognition result) for the speech recognition processing result information. Then, the process returns to step S202.

  (Step S207) The output unit 108 outputs information indicating that speech recognition has failed. Then, the process returns to step S202.

  In the flowchart of FIG. 2, the speech recognition process may be performed only when the pronunciation evaluation process is performed first and the pronunciation evaluation result has a predetermined relationship with the threshold value.

  Further, in the flowchart of FIG. 2, the unit for accepting voice is not limited to words, phrases, sentences, two or more sentences, and the like. Moreover, the unit for performing speech recognition and the unit for performing pronunciation evaluation are not limited. In the flowchart of FIG. 2, for example, one word speech is received, one word speech is recognized, and pronunciation evaluation of one word speech is performed. When the unit for receiving speech is a sentence, for example, the speech recognition process in step S203 and the pronunciation rating process in step S204 are repeatedly executed for the number of pronunciation intervals.

  Further, in the flowchart of FIG. 2, the processing is ended by powering off or interruption of processing end.

  Next, the pronunciation rating process in step S204 of FIG. 2 will be described using the flowchart of FIG.

  (Step S301) The frame classification unit 1061 classifies the voice received by the voice receiving unit 102 into frames. At this stage, frame audio data which is audio data for each divided frame is configured. The frame division processing performed by the frame classification unit 1061 is, for example, pre-processing when the frame audio data acquisition unit 1062 extracts frame audio data, and the input audio data is divided into all frames at once. Not exclusively.

  (Step S302) The frame audio data acquisition means 1062 substitutes 1 for the counter i.

  (Step S303) The frame audio data acquisition unit 1062 determines whether or not the i-th frame is present in the audio frames classified in step S301. If the i-th frame exists, the process goes to step S304, and if the i-th frame does not exist, the process goes to step S306.

  (Step S304) The frame sound data acquisition unit 1062 acquires the i-th frame sound data. The acquisition of frame sound data means, for example, performing sound analysis on the divided sound data, extracting feature vector data, and placing the data on a memory. Note that the frame audio data is, for example, data obtained by dividing the input audio data into frames. The frame audio data includes, for example, feature vector data extracted from the divided audio data by audio analysis. This feature vector data is, for example, MFCC obtained by discrete cosine transform of a filter bank output of 24 channels using a triangular filter, and the static parameter, delta parameter, and delta delta parameter are further normalized to 12 dimensions, respectively. Power and delta power and delta delta power (39th dimension).

  (Step S305) The frame audio data acquisition unit 1062 increments the counter i by 1. Then, the process returns to step S303.

  (Step S306) The pronunciation interval probability value acquisition unit 1064 calculates the forward likelihood and the backward likelihood of all states of all frames. Then, probability values for all frames and all states are obtained. Specifically, the pronunciation interval probability value acquisition unit 1064 calculates, for example, a posterior probability that each feature vector is generated from the target state. This posterior probability corresponds to the occupation frequency appearing in the Baum-Welch algorithm in the maximum likelihood estimation of the HMM. The Baum-Welch algorithm is a known algorithm and will not be described.

  (Step S307) The sounding section probability value acquisition unit 1064 calculates optimal state probability values of all frames.

  (Step S308) The pronunciation interval probability value acquisition unit 1064 reads all the probability values of one or more optimum states corresponding to the pronunciation interval (for example, a word interval).

  (Step S309) The rating value calculation means 1065 calculates a speech rating value using one or more probability values for each of one or more pronunciation sections acquired in step S308 as parameters. For example, an average value (time average value) of one or more probability values acquired in step S308 is calculated. Then, the process returns to the upper function. Details of the rating value calculation process will be described later.

  Hereinafter, a specific operation of the speech recognition apparatus in the present embodiment will be described.

First, in this speech recognition apparatus, a native pronunciation phoneme HMM is learned from a native pronunciation speech database of a language of a recognition target word (which may be a sentence) by means not shown. Here, the number of phoneme types is L, and the HMM for the l-th phoneme is λ _l . Since this learning process is a known technique, a detailed description thereof is omitted. The HMM specifications are shown in FIG. The specification of the HMM is the same in the description of specific examples in other embodiments. However, it goes without saying that the specifications of the HMM may be other specifications.

  Then, from the learned L types of phoneme HMMs, HMMs corresponding to one or more phonemes constituting speech such as words or sentences to be recognized are acquired, and the acquired one or more HMMs are connected in the order of phonemes. Audio data is constructed. The acoustic data is stored in the acoustic data storage unit 103.

  Next, a user (here, for example, an American) inputs a voice recognition start instruction. Such an instruction is made, for example, by pressing a predetermined button with a mouse.

  Next, the user pronounces the voice “right”. Then, the voice receiving unit 102 receives an input of a voice generated by the user.

  Next, it is assumed that the speech recognition unit 105 performs speech recognition processing to obtain a phoneme transcription “right” as a recognition result.

  Next, the pronunciation rating unit 106 performs a pronunciation rating as follows.

  That is, the frame classification unit 1061 classifies the voice received by the voice receiving unit 102 into short frames. It is assumed that the frame interval is determined in advance.

The frame audio data acquisition unit 1062 performs spectrum analysis on the audio data classified by the frame classification unit 1061 and calculates a feature vector series “O = o ₁ , o ₂ ,..., O _T ”. T is a sequence length. Here, the feature vector series is a set of feature vectors of each frame. The feature vector is, for example, an MFCC obtained by performing discrete cosine transform on a filter bank output of 24 channels using a triangular filter, and the static parameter, the delta parameter, and the delta delta parameter are further normalized to 12 dimensions, respectively. Power and delta power and delta delta power (39th dimension). In spectral analysis, it is preferable to perform cepstrum average removal. The voice analysis conditions are shown in the table of FIG. Note that the voice analysis conditions are the same in the description of specific examples in other embodiments. However, it goes without saying that the voice analysis conditions may be other conditions.

Then, the optimal state determination unit 1063, based on the feature vector o _t constituting the obtained feature vector series, to determine the optimal conditions for a given frame (optimum condition for the feature vector o _t). The algorithm by which the optimum state determination unit 1063 determines the optimum state is, for example, the Viterbi algorithm. In such a case, the optimum state determination unit 1063 determines the optimum state using the HMM connected as described above. The optimum state determination unit 1063 obtains an optimum state sequence that is an optimum state of two or more frames.

Next, the pronunciation interval probability value acquisition unit 1064 first calculates the optimum state probability value (γ _t (q _t ^* )) in the optimum state (q _t ^* ) by the following formula 1. Note that γ _t (q _t ^* ) is a value obtained by substituting q _t ^* for j in the posterior probability function γ _t (j) of the state j. Then, the posterior probability function γ _t (j) of the state j is calculated using Equation 2. This probability value _{(γ t} (j)) is, t th feature vector o _t is the posterior probability that is generated from the state j, is calculated using dynamic programming. Note that j is a state identifier for identifying a state.

In Equation 1, _{q t} represents the state identifier for _{o t.} This probability value (γ _t (j)) corresponds to the occupation frequency appearing in the Baum-Welch algorithm in the maximum likelihood estimation of the HMM.

  In Equation 2, “αt (j)” and “βt (j)” are calculated by the forward-backward algorithm using all the HMMs. “Αt (j)” is a forward likelihood, and “βt (j)” is a backward likelihood. Since the Baum-Welch algorithm and the forward-backward algorithm are known algorithms, detailed description thereof is omitted.

  In Equation 2, N represents the total number of states over all HMMs.

Then, the pronunciation interval probability value acquisition unit 1064 acquires all the probability values of one or more optimum states corresponding to the pronunciation interval. Then, the rating value calculating means 1065 calculates an average value (time average value) of the acquired one or more probability values. Specifically, the rating value calculation means 1065 calculates a rating value using Equation 3.

  Note that the pronunciation rating unit 106 may first obtain an optimum state, and then obtain a probability value of the optimum state (the probability value is 0 or more and 1 or less). First, the probability values of all states may be obtained, then the optimum state for each feature vector of the feature vector series may be obtained, and the probability value corresponding to the optimum state may be obtained.

  If the utterance corresponding to the user's t-th frame is close to the pronunciation indicated by the acoustic data (for example, correct native pronunciation), the numerator value of Equation (2) in Equation 2 is set to all other possible values. As a result, the probability value (rating value) in the optimum state becomes large. Conversely, if the section is not close to the pronunciation indicated by the acoustic data, the rating value is small. In the case where it is not close to any native pronunciation, the DAP rating value (also referred to as DAP score) is approximately equal to 1 / N. Since N is the number of all states in all phoneme HMMs, it is usually a large value and the DAP score is sufficiently small. Here, the DAP score is defined by the ratio between the probability value in the optimum state and the probability values in all possible states. Therefore, even if there is a slight spectrum variation due to differences in speaker characteristics and sound collection environment, if the user pronounces correctly, the variation is offset and a score with a high rating value is maintained.

  FIG. 6 and FIG. 7 show the DAP score (score for each frame) calculated by the rating value calculation means 1065 in the middle. 6 and 7, the horizontal axis represents the analysis frame number, and the vertical axis represents the score in%. A thick broken line represents a phoneme boundary, a thin dotted line represents a state boundary (both obtained by the Viterbi algorithm), and a phoneme name is shown at the top of the figure. FIG. 6 shows a DAP score for pronunciation of English “right” by an American male. The horizontal axis and vertical axis of the graph indicating the rating value are the same in the graph described later.

  FIG. 7 shows a DAP score of pronunciation of English “right” by a Japanese male. American pronunciation is basically higher than Japanese pronunciation. Also, in FIG. 6, it can be seen that the score has dropped in some places at the boundary of the state.

  Then, it is assumed that the rating value calculation means 1065 calculates, for example, a pronunciation rating result (also referred to as a t-DAP score) in the pronunciation interval as “68” (see FIG. 8).

  Next, the determination unit 107 compares the threshold value “50” stored in advance with the pronunciation evaluation result “68” to determine whether or not a predetermined condition is satisfied. The predetermined condition is, for example, “pronunciation rating result> threshold”. Since “68> 50”, the determination unit 107 determines that a predetermined condition is satisfied.

  Next, the output unit 108 outputs the voice recognition processing result information “right” acquired by the voice recognition unit 105. Such output is, for example, display on a display.

  Next, another user (assuming they are Japanese) pronounces the voice “right”. Then, the voice receiving unit 102 receives an input of a voice generated by the user.

  Next, it is assumed that the speech recognition unit 105 performs speech recognition processing and obtains a phoneme transcription “light” as a recognition result.

  Next, the pronunciation rating unit 106 performs pronunciation rating using the algorithm described above. It is assumed that the pronunciation evaluation result “29” is obtained.

  Next, the determination unit 107 compares the threshold value “50” stored in advance with the pronunciation evaluation result “29” to determine whether or not a predetermined condition is satisfied. Since “50> 29”, the determination unit 107 determines that the predetermined condition is not satisfied.

  Then, the output unit 108 outputs prestored information “speech recognition failed”. Such output is, for example, audio output from a speaker and display on a display.

  As described above, according to the present embodiment, it is possible to provide a speech recognition apparatus that outputs a speech recognition result only when a pronunciation rating is performed and a predetermined rating result is obtained (only when it is good). With this process, the possibility of outputting an erroneous recognition result can be reduced, and a highly reliable speech recognition apparatus can be provided. In addition, the user can use the device with peace of mind without impairing the reliability of the user with respect to the voice recognition device.

  It should be noted that according to the present embodiment, after performing the speech recognition processing, the pronunciation evaluation is performed. However, after performing the pronunciation evaluation, the speech recognition process may be performed only when the evaluation result satisfies a predetermined condition. The same applies to other embodiments.

  Moreover, according to this Embodiment, the algorithm of the speech recognition which a speech recognition part performs is not ask | required. In speech recognition, phoneme transcription in the recognition candidate data storage unit 104 may or may not be used. This also applies to other embodiments.

  Further, according to the present embodiment, the rating value calculation means for performing pronunciation evaluation calculates the time average value of the optimum state probability values in one or more optimum states. However, the rating value calculation means is a function having an optimum state probability value in one or more optimum states as a parameter, and any function can be used to calculate the score as long as an answer indicating good or bad pronunciation is obtained. May be. For example, the rating value calculation means may obtain a median value and 10 values before and after the median value among the optimum state probability values in one or more optimum states, and use the average value of the 11 values as a score.

  Also, according to the present embodiment, if the score of part of the sentence is lower than the threshold as a result of pronunciation evaluation, the recognition result of the whole sentence is not output, for example, the recognition result of the whole sentence is not output. Not always.

  Furthermore, the processing in the present embodiment may be realized by software. Then, this software may be distributed by software download or the like. Further, this software may be recorded on a recording medium such as a CD-ROM and distributed. This also applies to other embodiments in this specification. Note that the software that implements the information processing apparatus according to the present embodiment is the following program. In other words, this program uses a voice input step for receiving voice input to a computer, and performs voice recognition processing on the voice received in the voice input step using stored acoustic data. A speech recognition step for acquiring speech recognition processing result information, and a pronunciation rating step for performing pronunciation rating processing using the acoustic data for the speech received in the speech input step and acquiring a pronunciation rating result And a speech recognition acquired in the speech recognition step only when it is determined in the determination step whether the pronunciation rating result has a predetermined relationship with a predetermined threshold, and in the determination step. A program for executing an output step of outputting a speech recognition result for processing result information.

  In the above program, the pronunciation rating step performs a pronunciation rating process on the voice received in the voice input step by using the voice recognition processing result information acquired in the voice recognition step and the acoustic data. It is preferable to obtain a rating result.

  Further, in the above program, the pronunciation rating step includes a frame segmentation step for segmenting the speech accepted in the speech acceptance step into frames, and a frame for obtaining one or more frame speech data that is speech data for each of the segmented frames. An audio data acquisition step, an optimal state determination step for determining an optimal state of the one or more frame audio data, and a probability value of the optimal state determined in the optimal state determination step for each of the pronunciation intervals It is preferable to include an acquisition step and a rating value calculation step of calculating a voice rating value using one or more probability values for each of one or more pronunciation intervals acquired in the pronunciation interval probability value acquisition step as parameters. .

(Embodiment 2)
The speech recognition apparatus according to the present embodiment is different from the speech recognition apparatus according to the first embodiment in the pronunciation rating algorithm in the pronunciation rating unit. In the present embodiment, the rating value is calculated by evaluating the probability values of all the states in the phoneme including the optimum state in the pronunciation interval. The posterior probability calculated by the pronunciation rating device in the present embodiment is called tp-DAP with respect to t-DAP described in the first embodiment.

  FIG. 9 is a block diagram of the speech recognition apparatus in the present embodiment.

  The speech recognition apparatus in FIG. 9 differs from the speech recognition apparatus in FIG. 1 only in the pronunciation rating unit 906.

  The pronunciation rating unit 906 includes frame classification means 1061, frame sound data acquisition means 1062, optimum state determination means 1063, pronunciation interval frame phoneme probability value acquisition means 9061, and rating value calculation means 9062.

  The pronunciation interval frame phoneme probability value acquisition unit 9061 constituting the pronunciation evaluation unit 906 acquires, for each pronunciation interval, one or more probability values in the entire phoneme state having the optimal state of each frame determined by the optimal state determination unit 1063. To do.

  The rating value calculation means 9062 calculates a speech rating value using one or more probability values for each of one or more pronunciation intervals acquired by the pronunciation interval frame phoneme probability value acquisition means 9061 as parameters. For example, the rating value calculating unit 9062 obtains, for each frame, a sum of one or more probability values in the entire phonological state having the optimal state of each frame determined by the optimal state determining unit 1063, and the probability value for each frame. 1 or more is obtained based on the sum of the above, and the speech rating value is calculated using the one or more time average values as parameters.

  The pronunciation interval frame phoneme probability value acquisition means 9061 and the rating value calculation means 9062 can be usually realized by an MPU, a memory, or the like. The processing procedure of the pronunciation interval frame phoneme probability value acquisition means 9061 and the like is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  Next, the operation of this speech recognition apparatus will be described. This speech recognition apparatus differs from the speech recognition apparatus according to the first embodiment only in the pronunciation rating process in step S204. The pronunciation rating process of the speech recognition apparatus will be described with reference to the flowchart of FIG. In the flowchart of FIG. 10, only steps different from those in FIG. 3 will be described.

  (Step S1001) The pronunciation interval frame phoneme probability value acquisition unit 9061 substitutes 1 for j.

  (Step S1002) The sounding section frame phoneme probability value acquisition unit 9061 determines whether or not the jth frame is present in the main sounding section (for example, a word). If the j-th frame exists, the process goes to step S1003, and if the j-th frame does not exist, the process jumps to step S1006.

  (Step S1003) The pronunciation period frame phoneme probability value acquisition unit 9061 acquires all probability values of phonemes including the optimal state of the j-th frame.

  (Step S1004) The rating value calculation means 9062 calculates a rating value of one frame of speech using one or more probability values acquired in step S1003 as a parameter, and temporarily stores it in the memory.

  (Step S1005) The sounding section frame phoneme probability value acquisition unit 9061 increments j by 1. Then, the process returns to step S1002.

  (Step S1006) The rating value calculation means 9062 calculates the rating value of the main pronunciation section. For example, the rating value calculation unit 9062 obtains, for each frame, the sum of one or more probability values in the entire phoneme state having the optimal state of each frame determined by the optimal state determination unit 1063, and the probability value for each frame. Based on the sum of the above, the time average value of the sum of the probability values of the sounding section is calculated as a speech evaluation value of the sounding section. Then, the process returns to the upper function.

  Hereinafter, a specific operation of the speech recognition apparatus in the present embodiment will be described. In the present embodiment, the rating value calculation algorithm is different from that of the first embodiment, and therefore the operation will be mainly described.

  First, after the user inputs a start instruction, the user utters a speech to be recognized (here, “right”). Then, the voice receiving unit 102 receives an input of a voice generated by the user.

  Next, the frame classification unit 1061 classifies the voice received by the voice receiving unit 102 into short frames.

The frame audio data acquisition unit 1062 performs spectrum analysis on the audio data classified by the frame classification unit 1061 and calculates a feature vector series “O = o ₁ , o ₂ ,..., O _T ”.

  Next, the pronunciation interval frame phoneme probability value acquisition unit 9061 calculates the posterior probability (probability value) of each state of each frame. The probability value can be calculated by the above-described Equations 1 and 2.

Then, the optimal state determination unit 1063, based on the feature vector o _t constituting the obtained feature vector series to determine the optimum conditions for each frame (optimum condition for the feature vector o _t). That is, the optimum state determination unit 1063 obtains an optimum state sequence. The order of calculating the posterior probability (probability value) of each state in each frame and determining the optimum state is not limited.

  Next, the pronunciation interval frame phoneme probability value acquisition unit 9061 acquires, for each pronunciation interval, all probability values of phonemes including the optimum state of each frame included in the pronunciation interval.

Then, the rating value calculation means 9062 calculates the sum of all probability values of phonemes including the optimal state of each frame for each frame. First, the rating value calculation unit 9062 calculates the sum of all the probability values of the phoneme including the optimal state of the frame (this is referred to as “p-DAP”) by the following Equation 4.

Then, the rating value calculation means 9062 averages the sum of the probability values calculated for each frame for each sounding section, and calculates a rating value for each sounding section. Specifically, the rating value calculation unit 9062 calculates the rating value by Equation 5.

  The rating values (also referred to as “tp-DAP score”) calculated by the rating value calculating means 9062 are shown in the table of FIG. FIG. 11 shows the evaluation results of an American male and a Japanese male. Phoneme and Word indicate the range of time average in tp-DAP. Here, a time average of p-DAP is adopted instead of DAP in the first embodiment. In FIG. 11, the pronunciation value of an American male is higher than the pronunciation value of a Japanese male, and a good evaluation result is obtained.

  Next, the determination unit 107 compares the threshold value “60” stored in advance with the pronunciation evaluation result “84” in the case of an American male utterance, and determines whether or not a predetermined condition is satisfied. Since “84> 60”, the determination unit 107 determines that a predetermined condition is satisfied.

  Next, the output unit 108 outputs the voice recognition processing result information “right” acquired by the voice recognition unit 105.

  Further, the determination unit 107 compares the threshold value “60” stored in advance with the pronunciation evaluation result “33” in the case of a Japanese male utterance, and determines that the predetermined condition is not satisfied.

  Then, the output unit 108 outputs prestored information “speech recognition failed”.

  As described above, according to the present embodiment, it is possible to provide a speech recognition apparatus that outputs a speech recognition result only when a pronunciation rating is performed and a predetermined rating result is obtained (only when it is good). With this process, the possibility of outputting an erroneous recognition result can be reduced, and a highly reliable speech recognition apparatus can be provided. In addition, the user can use the device with peace of mind without impairing the reliability of the user with respect to the voice recognition device. Furthermore, according to the present embodiment, pronunciation evaluation can be performed with high accuracy, and as a result, a correct recognition result can be output with high accuracy.

  Note that according to the present embodiment, the rating value calculation means for performing pronunciation rating calculates the time average value of the sum of one or more probability values in the entire phoneme state having the optimal state of each frame. However, the rating value calculation means is a function that uses as a parameter the sum of one or more probability values in the overall state of the phoneme having the optimum state of each frame, and if it is a function that can obtain an answer indicating good or bad pronunciation, The score may be calculated by any function. For example, the rating value calculation means may obtain a median value and a value before and after the median value of the sum of one or more probability values in the entire phoneme state, and use the average value of the 31 values as a score. .

  Furthermore, the software that implements the information processing apparatus according to the present embodiment is the following program. In other words, this program uses a voice input step for receiving voice input to a computer, and performs voice recognition processing on the voice received in the voice input step using stored acoustic data. A speech recognition step for acquiring speech recognition processing result information, and a pronunciation rating step for performing pronunciation rating processing using the acoustic data for the speech received in the speech input step and acquiring a pronunciation rating result And a speech recognition acquired in the speech recognition step only when it is determined in the determination step whether the pronunciation rating result has a predetermined relationship with a predetermined threshold, and in the determination step. A program for executing an output step of outputting a speech recognition result for processing result information.

  In the above program, the pronunciation rating step performs a pronunciation rating process on the voice received in the voice input step by using the voice recognition processing result information acquired in the voice recognition step and the acoustic data. It is preferable to obtain a rating result.

  Further, in the above program, the pronunciation rating step includes a frame division step for dividing the sound received by the sound receiving unit into frames, an optimum state determination step for determining an optimum state of the one or more frame sound data, A pronunciation interval frame phoneme probability value acquisition step of acquiring, for each pronunciation interval, one or more probability values in the overall state of the phoneme having the optimal state of each frame determined in the optimal state determination step; and the pronunciation interval frame phoneme probability value It is preferable to include a rating value calculation step of calculating a voice rating value using one or more probability values for each of one or more pronunciation intervals acquired in the acquisition step as a parameter.

(Embodiment 3)
The speech recognition apparatus in the present embodiment is different from the speech recognition apparatus in Embodiments 1 and 2 in the pronunciation rating algorithm in the pronunciation rating unit. In the present embodiment, the speech recognition apparatus is an apparatus capable of detecting a silent section and performing pronunciation evaluation considering the silent section. More specifically, the speech recognition apparatus normally performs pronunciation evaluation without considering frames in silent intervals.

  FIG. 12 is a block diagram of the speech recognition apparatus in the present embodiment.

  The speech recognition apparatus in FIG. 12 differs from the speech recognition apparatus in FIG. 1 only in the pronunciation rating unit 1206.

  The pronunciation rating unit 1206 includes frame classification means 1061, frame audio data acquisition means 1062, optimum state determination means 1063, pronunciation interval probability value acquisition means 1064, silence data storage means 12061, silence interval detection means 12062, and rating value calculation means 12063. It has.

  The pronunciation rating unit 1206 performs a pronunciation rating process on the voice received by the voice receiving unit 102 using the acoustic data in the acoustic data storage unit 103, and acquires a pronunciation rating result. In general, the pronunciation rating unit 106 performs a pronunciation rating process on the voice received by the voice receiving unit 102 using the voice recognition processing result information and the acoustic data acquired by the voice recognition unit 105, and obtains the pronunciation rating result. get.

  The pronunciation rating unit 1206 can be usually realized by an MPU, a memory, or the like. The processing procedure of the pronunciation rating unit 1206 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The silence data storage unit 12061 constituting the pronunciation rating unit 1206 stores silence data indicating silence. The silence data is preferably data based on HMM in which hidden Markov models for each phoneme are connected. Further, the silence data is not necessarily data based on the HMM. The silence data may be data based on other models such as a single Gaussian distribution model, a probability model (GMM: Gaussian mixture model), and a statistical model. The silent data storage unit 12061 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium.

  The silent section detecting unit 12062 detects a silent section based on the input voice data and the silent data stored in the silent data storage unit 12061. More specifically, the silent section detecting unit 12062 detects a silent section based on the frame voice data acquired by the frame voice data acquiring unit 1062 and the silent data of the silent data storage unit 12061. When the similarity between the frame audio data acquired by the frame audio data acquisition unit 1062 and the silence data is equal to or greater than a predetermined value, the silence interval detection unit 12062 determines that the frame audio data is data of the silence interval. Also good. Further, the silent section detecting means 12062 has a frame probability value of each frame acquired by the sounding section probability value acquiring means 1064 described below below a predetermined value and the frame sound data acquired by the frame sound data acquiring means 1062 When the similarity of silence data is greater than or equal to a predetermined value, it may be determined that the frame audio data is data in a silent section.

  The rating value calculation means 12063 evaluates speech by using one or more probability values for one or more sounding sections acquired by the sounding section probability value acquisition means 1064 as parameters, except for the silent section detected by the silent section detection means 12062. Calculate the value. Note that it does not matter how the rating value calculation means 12063 calculates the rating value using the probability value. Here, for example, it is assumed that the rating value calculation unit 12063 calculates the rating value by t-DAP. The rating value calculation means 12063 does not necessarily calculate the rating value except for the silent section. The rating value calculation means 12063 may calculate the rating value so as to reduce the influence of the silent section.

  The silent section detection means 12062 and the rating value calculation means 12063 can be usually realized by an MPU, a memory, or the like. The processing procedure of the silent section detecting means 12062 and the like is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  Next, the operation of this speech recognition apparatus will be described. This speech recognition apparatus differs from the speech recognition apparatus according to the first embodiment only in the pronunciation rating process in step S204. The pronunciation rating process of this speech recognition apparatus will be described using the flowchart of FIG. In the flowchart of FIG. 13, only steps different from those in FIG. 3 will be described.

  (Step S1301) The pronunciation interval probability value acquisition unit 1064 substitutes 1 for the counter i.

  (Step S1302) The sounding section probability value acquisition unit 1064 determines whether or not the i-th frame sound data exists. If the i-th frame sound data exists, the process proceeds to step S1303. If the i-th frame sound data does not exist, the process proceeds to step S1309. Note that the i-th frame audio data is the i-th data in the frame audio data acquired in step S304.

  (Step S1303) The silent section detecting means 12062 acquires the optimal state probability value of the i-th frame sound data.

  (Step S1304) The silent section detection unit 12062 determines whether or not the probability value acquired in step S1303 is lower than (or lower than) the threshold value. If it is lower than the threshold value (or lower than the threshold value), go to step S1305, and if it is lower than the threshold value (or lower than the threshold value), go to step S1307.

  (Step S1305) The silent section detecting unit 12062 acquires the silent data of the silent data storage unit 12061 and all the acoustic data of the acoustic data storage unit 103.

  (Step S1306) The silent section detecting means 12062 determines whether or not the i-th frame audio data has the highest probability value of silent data. If the silence data has the highest probability value, the process proceeds to step S1308. If the silence data has the highest probability value, the process proceeds to step S1307.

  (Step S1307) The pronunciation interval probability value acquisition means 1064 temporarily stores the probability value acquired in Step S1303.

  (Step S1308) The sounding section probability value acquisition unit 1064 increments the counter i by 1, and goes to step S1302.

  (Step S1309) The rating value calculation means 12063 calculates a voice rating value using the one or more probability values temporarily stored in step S1307 as parameters, and returns to the upper function.

  In the flowchart of FIG. 13, it is preferable that the rating value calculation unit 12063 calculates the score while ignoring the rating value of the silent section as described above. For example, the score may be calculated with 1/10.

  Hereinafter, a specific operation of the speech recognition apparatus in the present embodiment will be described. In the present embodiment, since the rating value is calculated in consideration of the silent section, the rating value calculation algorithm is different from that of the first embodiment. Therefore, the different processing will be mainly described.

  First, the user inputs a start instruction. Next, for example, the user pronounces the speech to be recognized. Then, the voice receiving unit 102 receives an input of a voice generated by the user.

  Next, the frame classification unit 1061 classifies the voice received by the voice receiving unit 102 into short frames.

The frame audio data acquisition unit 1062 performs spectrum analysis on the audio data classified by the frame classification unit 1061 and calculates a feature vector series “O = o ₁ , o ₂ ,..., O _T ”.

Then, the optimal state determination unit 1063, based on the feature vector o _t constituting the obtained feature vector series, to determine the optimal conditions for a given frame (optimum condition for the feature vector o _t).

Next, the pronunciation interval probability value acquisition unit 1064 calculates the forward likelihood and the backward likelihood of all states of all frames. Then, the sounding section probability value acquisition unit 1064 calculates the optimum state probability value (γ _t (q _t ^* )) in the optimum state for each frame using Equation 1 for each frame.

  Furthermore, the silent section detection means 12062 acquires the probability value of the optimal state for each frame audio data, and determines whether or not the acquired probability value is lower than a predetermined value.

  Then, when the silence section detection means 12062 determines that the acquired probability value is lower than a predetermined value, the value of the posterior probability (DAP score) indicating the similarity between the frame sound data and the silence data is determined as the frame sound. It is determined whether or not the value is higher than any of the posterior probability values indicating the similarity between the data and other data. If the value of the posterior probability indicating the similarity between the frame sound data and the silence data is the highest, the pronunciation rating unit 1206 ignores the frame as a frame in the silence interval. Then, the rating value calculation means 12063 calculates a t-DAP score excluding the silent data section. Here, it is assumed that the rating value calculating unit 12063 calculates “75” and the t-DAP score.

  Next, the determination unit 107 compares the threshold value “60” stored in advance with the pronunciation evaluation result “75” to determine whether or not a predetermined condition is satisfied. Since “75> 60”, the determination unit 107 determines that a predetermined condition is satisfied.

  Next, the output unit 108 outputs the speech recognition processing result information (for example, “right”) acquired by the speech recognition unit 105.

  Such output processing is the same as the processing in the first embodiment.

  As described above, according to the present embodiment, it is possible to provide a speech recognition apparatus that outputs a speech recognition result only when a pronunciation rating is performed and a predetermined rating result is obtained (only when it is good). With this process, the possibility of outputting an erroneous recognition result can be reduced, and a highly reliable speech recognition apparatus can be provided. In addition, the user can use the device with peace of mind without impairing the reliability of the user with respect to the voice recognition device. Furthermore, according to the present embodiment, the speech recognition result is output based on the result of pronunciation evaluation excluding the silent section, and is not output. Therefore, the correct recognition result can be output with higher accuracy. .

  It should be noted that according to the present embodiment, the pronunciation rating algorithm in the rating value calculating means is not limited. The pronunciation rating algorithm may be t-DAP, tp-DAP, or other algorithms.

  Furthermore, the software that implements the information processing apparatus according to the present embodiment is the following program. In other words, this program uses a voice input step for receiving voice input to a computer, and performs voice recognition processing on the voice received in the voice input step using stored acoustic data. A speech recognition step for acquiring speech recognition processing result information, and a pronunciation rating step for performing pronunciation rating processing using the acoustic data for the speech received in the speech input step and acquiring a pronunciation rating result And a speech recognition acquired in the speech recognition step only when it is determined in the determination step whether the pronunciation rating result has a predetermined relationship with a predetermined threshold, and in the determination step. A program for executing an output step of outputting a speech recognition result for processing result information.

  Further, in the pronunciation rating step of the program, the program further includes a silent section detecting step for detecting a silent section based on the input voice data and the stored silent data, and the optimum state determining step includes the silent state determining step. It is preferable to determine the optimum state of the one or more frame audio data excluding the frame audio data for the section.

  In each of the above embodiments, each process (each function) may be realized by centralized processing by a single device (system), or by distributed processing by a plurality of devices. May be.

  FIG. 14 shows the external appearance of a computer that executes the programs described in this specification to realize the speech recognition apparatuses according to the various embodiments described above. The above-described embodiments can be realized by computer hardware and a computer program executed thereon. FIG. 14 is an overview diagram of the computer system 340, and FIG. 15 is a block diagram of the computer system 340.

  In FIG. 14, a computer system 340 includes a computer 341 including an FD (Flexible Disk) drive and a CD-ROM (Compact Disk Read Only Memory) drive, a keyboard 342, a mouse 343, a monitor 344, and a microphone 345. .

  In FIG. 15, in addition to the FD drive 3411 and the CD-ROM drive 3412, the computer 341 includes a CPU (Central Processing Unit) 3413, a bus 3414 connected to the CPU 3413, the CD-ROM drive 3412, and the FD drive 3411, and a boot. A ROM (Read-Only Memory) 3415 for storing a program such as an up program, and a RAM (Random Access Memory) connected to the CPU 3413 for temporarily storing instructions of an application program and providing a temporary storage space 3416 and a hard disk 3417 for storing application programs, system programs, and data. Although not shown here, the computer 341 may further include a network card that provides connection to the LAN.

  A program that causes the computer system 340 to execute the functions of the sound processing apparatus according to the above-described embodiment is stored in the CD-ROM 3501 or the FD 3502, inserted into the CD-ROM drive 3412 or the FD drive 3411, and further stored in the hard disk 3417. May be forwarded. Alternatively, the program may be transmitted to the computer 341 via a network (not shown) and stored in the hard disk 3417. The program is loaded into the RAM 3416 at the time of execution. The program may be loaded directly from the CD-ROM 3501, the FD 3502, or the network.

  The program does not necessarily include an operating system (OS), a third-party program, or the like that causes the computer 341 to execute the functions of the voice processing apparatus according to the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 340 operates is well known and will not be described in detail.

  Further, the computer that executes the program may be singular or plural. That is, centralized processing may be performed, or distributed processing may be performed.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the speech recognition apparatus according to the present invention has an effect of preventing an erroneous recognition result from being output, and is useful as a high-performance speech recognition apparatus or the like.

Block diagram of speech recognition apparatus according to Embodiment 1 Flow chart for explaining the operation of the voice recognition apparatus Flow chart for explaining the pronunciation evaluation process Diagram showing the specifications of the HMM Figure showing the same voice analysis conditions Figure showing the DAP score Figure showing the DAP score The figure which shows the output example of the t-DAP score Block diagram of a speech recognition apparatus according to Embodiment 2 A flowchart for explaining the operation of the pronunciation evaluation process The figure which shows the same tp-DAP score Block diagram of speech recognition apparatus according to Embodiment 3 A flowchart for explaining the operation of the pronunciation evaluation process External view of a computer that implements the speech recognition apparatus Block diagram of a computer system for realizing the voice recognition apparatus

Explanation of symbols

DESCRIPTION OF SYMBOLS 101 Reception part 102 Voice reception part 103 Acoustic data storage part 104 Recognition candidate data storage part 105 Speech recognition part 106,906,1206 Pronunciation evaluation part 107 Judgment part 108 Output part 1061 Frame division | segmentation means 1062 Frame audio | voice data acquisition means 1063 Optimal state determination Means 1064 Pronunciation interval probability value acquisition means 1065, 9062, 12063 Rating value calculation means 9061 Pronunciation interval frame phoneme probability value acquisition means 12061 Silence data storage means 12062 Silence interval detection means

Claims (7)

An acoustic data storage unit that stores acoustic data that is data relating to the target speech to be compared;
A voice input unit that accepts voice input;
A voice recognition unit that performs voice recognition processing on the voice received by the voice input unit using the acoustic data, and acquires voice recognition processing result information that is a result of the voice recognition processing;
A pronunciation rating unit that performs pronunciation rating processing using the acoustic data and obtains a pronunciation rating result for the voice received by the voice input unit;
A determination unit that determines whether or not the pronunciation rating result has a predetermined relationship with a predetermined threshold;
A speech recognition apparatus comprising: an output unit that outputs a speech recognition result for speech recognition processing result information acquired by the speech recognition unit only when the determination unit determines that there is a predetermined relationship. The pronunciation rating unit
The speech recognition according to claim 1, wherein the speech received by the speech input unit is subjected to pronunciation rating processing using the speech recognition processing result information acquired by the speech recognition unit and the acoustic data, and the pronunciation rating result is acquired. apparatus. The pronunciation rating unit
Frame dividing means for dividing the sound received by the sound receiving unit into frames;
Frame audio data acquisition means for obtaining one or more frame audio data which is audio data for each of the divided frames;
Optimal state determining means for determining an optimal state of the one or more frame audio data;
A pronunciation interval probability value acquisition means for acquiring the probability value of the optimum state determined by the optimum state determination means for each pronunciation interval;
The rating value calculation means which calculates the rating value of a voice using one or more probability values for every one or more pronunciation sections acquired by the pronunciation section probability value acquisition means as a parameter. Voice recognition device. The pronunciation rating unit
Frame dividing means for dividing the sound received by the sound receiving unit into frames;
Optimal state determining means for determining an optimal state of the one or more frame audio data;
A pronunciation interval frame phoneme probability value acquisition unit that acquires, for each pronunciation interval, one or more probability values in the overall state of the phoneme having the optimal state of each frame determined by the optimal state determination unit;
The rating value calculation means which calculates the rating value of a voice by using one or more probability values for each of one or more pronunciation sections acquired by the pronunciation section frame phoneme probability value acquisition means as a parameter. Voice recognition device. The pronunciation rating unit
Silence data storage means for storing silence data that is data indicating silence; and
Based on the input voice data and the silent data, further comprising a silent section detecting means for detecting a silent section,
The optimum state determining means includes
The speech recognition apparatus according to claim 3 or 4, wherein an optimum state of the one or more frame speech data excluding the frame speech data for the silent section is determined. The speech recognition apparatus according to any one of claims 1 to 5, wherein the acoustic data is data relating to a target speech to be compared, and is data based on an HMM in which hidden Markov models for each phoneme are connected. On the computer,
A voice input step for receiving voice input;
A voice recognition step of performing voice recognition processing on the voice received in the voice input step using the stored acoustic data, and acquiring voice recognition processing result information as a result of the voice recognition processing;
For the voice received in the voice input step, a pronunciation rating process is performed using the acoustic data to obtain a pronunciation rating result; and
A determination step of determining whether the pronunciation rating result is in a predetermined relationship with a predetermined threshold;
A program for executing an output step of outputting a speech recognition result for speech recognition processing result information acquired in the speech recognition step only when it is determined in the determination step that a predetermined relationship exists.