JPH1185185A - Voice recognition system and storage medium with voice recognition control program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the voice recognition rate by employing a smaller amount of processes under such an acoustic environment that different kinds of plural noise signals exist. SOLUTION: In a cancel circuit 31, digital signals 211, which are generated by capturing car audio signals in a noise input circuit 21, are eliminated from voice signals 101 caught by a voice input circuit 10, voice is recognized in a recognition circuit 40 by a hidden Markov model(HMM) method and a recognition word S1 and a recognition probability P1 are obtained. Similarly, in a cancel circuit 32, digital signals 221, which are noise signals outside the car caught by a noise input circuit 22, are eliminated from the signals 101, voice is recognized in the circuit 40 by the HMM method and a recognition word S2 and a recognition probability P2 are obtained. Moreover, in a cancel circuit 33, digital signals 231, which are engine sound signals caught by a noise input circuit 23, are eliminated from the signals 101, voice is recognized in a recognition circuit 40 by the HMM method and a recognition symbol S3 and a recognition probability P3 are obtained. Then, a comparison circuit 50 selects the recognized word having the highest probability among the recognition probabilities Ps.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a speech recognition system and a recording medium on which a speech recognition control program is recorded, for example, when a main sound signal contains noise signals from a plurality of noise sources. The present invention relates to a system for performing speech recognition by removing a noise signal from a signal.

[0002]

2. Description of the Related Art In recent years, it has been attempted to control the operation of a product such as a car navigation system by a user's voice using voice recognition. However, in a car where the noise is intense, such as the sound from the car audio system in the car and the noise from outside the car, the recognition rate is low and practical if the speech recognition device is applied as it is in a car. is not.

[0003] Therefore, conventionally, an adaptive noise canceller is used to reduce a noise signal superimposed on a car audio signal or the like from an audio signal captured using a microphone or the like, and thereafter, speech recognition is performed. It has been done.

[0004]

However, the noise signal is not limited to a car audio signal, but also includes a large number of different types of noise signals such as a noise outside a vehicle, a sound of cutting a wind, an engine sound, and a tire running sound. To cope with such a noise signal in which a plurality of types of noise exist, an adaptive noise canceller having many inputs is required. However, the multi-input adaptive noise canceller has been considered to be impractical due to the complicated algorithm and the enormous amount of calculation and the processing time.

[0005] For this reason, in an acoustic environment in which a plurality of different types of noise signals such as noise from the outside of a car and an acoustic signal from a car audio system are present, speech is processed with a realistic processing amount (small processing amount). There is a demand for a speech recognition system capable of improving the recognition accuracy of recognition and a recording medium recording a speech recognition control program.

[0006]

Therefore, according to the present invention, there is provided a main sound signal capturing means capable of capturing main sound and outputting a main sound signal and capturing noise from at least two noise sources. A speech recognition system is a system including at least two or more noise signal capturing means and adaptive noise canceling means for removing a noise signal from at least two or more noise sources, wherein the noise from one noise source is removed. First noise signal capturing means for capturing and outputting a noise capturing signal, second noise signal capturing means for capturing noise from another noise source and outputting a noise capturing signal, and first noise from the main acoustic signal First adaptive noise canceling means for removing the noise capturing signal captured by the signal capturing means and outputting the removed main acoustic signal; and noise captured by the second noise signal capturing means from the main acoustic signal. A second adaptive noise canceling means for removing the captured signal and outputting the removed main acoustic signal; and a speech recognition of the main acoustic signal output from the first adaptive noise canceling means by using a statistical acoustic model. The word and the likelihood of its recognition are obtained, and the main sound signal output from the second adaptive noise canceling means is also subjected to speech recognition using the statistical acoustic model, to obtain the recognition word and to obtain the certainty of the recognition. A speech recognition unit that outputs a recognition word having a high probability of speech recognition from these recognition results as a recognition result.

[0007] By adopting such a configuration, a noise capturing signal obtained by capturing noise from a certain noise source can be removed from the main acoustic signal by the first adaptive noise canceling means. Can be removed by the second adaptive noise canceling means, speech recognition is performed on each of the removed main sound signals, and a word having a high probability of recognition is determined as a recognition result. For this reason, even when noise signals from multiple noise sources are included in the main acoustic signal, speech recognition can be performed with a simple configuration without performing complicated processing, and a statistical acoustic model can be created. Since the speech recognition is used to select a recognized word having a high recognition rate, the recognition accuracy can be improved.

In the above-described configuration, two adaptive noise canceling means are used. This is because there are at least two types of noise signals from the noise source. If three or more capture units and three or more adaptive noise cancellation units are provided, speech recognition processing can be performed by similar processing.

Further, according to the present invention, a voice recognition system including main sound signal capturing means capable of capturing main sound and outputting a main sound signal and capturing noise from at least two or more noise sources is provided. A system including at least two or more noise signal capturing means and adaptive noise canceling means for removing noise signals from two or more noise sources, capturing noise from one noise source and generating a noise capturing signal. First noise signal capturing means for outputting, second noise signal capturing means for capturing noise from another noise source and outputting a noise capturing signal, and first noise signal capturing means for capturing the main acoustic signal. First adaptive noise canceling means for removing the noise capture signal and outputting the removed main audio signal, and removing the noise capture signal captured by the second noise signal capture means from the main audio signal A second adaptive noise canceling means for outputting the removed main sound signal, a signal quality of the output main sound signal of the first adaptive noise canceling means and a signal quality of the main sound signal output of the second adaptive noise canceling means, Voice recognition means for selecting an output main sound signal with good signal quality and performing voice recognition on the selected output main sound signal.

By adopting such a configuration, the main sound signal output from the first adaptive noise canceling unit and the main sound output from the second adaptive noise canceling unit are determined based on the signal quality from the signal-to-noise ratio or the amount of signal distortion. Since the signal quality of the signal can be determined and speech recognition is performed on the output main sound signal of the adaptive noise canceling unit having good signal quality, the recognition accuracy can be improved and the calculation processing amount of the speech recognition can be reduced. Can reduce the time for recognition.

Further, according to the present invention, the main sound captured by the main sound signal capturing means capable of capturing the main sound and outputting the main sound signal by the computer and capturing the noise from at least two or more noise sources. At least 2
A first noise signal capturing unit including at least two or more noise signal capturing units for capturing a noise signal from the above noise source, capturing noise from a certain noise source, and outputting a noise capturing signal; and Speech recognition for removing a noise capture signal captured by a second noise signal capture unit that captures noise from another noise source and outputs a noise capture signal, and performs voice recognition on the removed main acoustic signal The voice recognition control program of the recording medium on which the control program is recorded includes at least two or more adaptive noise canceling steps for removing at least two or more noise capture signals from the main sound signal, and includes a step of removing the main sound signal from the main sound signal. A first adaptive noise canceling step of removing the noise capture signal captured by the first noise signal capture means and outputting the removed main sound signal; A second adaptive noise canceling step of removing the noise capture signal captured by the second noise signal capturing means from the audio signal and outputting the removed main audio signal, and an output main sound of the first adaptive noise canceling step Let the signal be speech-recognized by the statistical acoustic model, ask for the recognition word and ask for the certainty of the recognition,
The output main acoustic signal of the second adaptive noise canceling step is also subjected to speech recognition using a statistical acoustic model to determine a recognized word and to determine the likelihood of the recognition. From these recognition results, the likelihood of the speech recognition is high. A speech recognition step of outputting a recognition word as a recognition result.

By adopting such a configuration, speech recognition can be performed with a simple configuration without performing complicated program processing, and speech recognition is performed using a statistical acoustic model, and a recognized word having a high recognition rate is obtained. , The recognition accuracy can be improved, and the voice recognition system can be reduced in size by storing it in a ROM, a flash memory, or a magnetic disk device.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, a preferred embodiment of the present invention will be described with reference to the drawings. In the present embodiment, in a speech recognition system that combines an adaptive noise canceller to reduce noise signals included in a speech signal, a plurality of noise signals are input, and an adaptive noise canceller unit corresponding to each noise input unit is provided. Adaptive noise cancellers and Hidden Markov model, a statistical acoustic model
A means for achieving a high recognition rate in cooperation with a speech recognition processing unit based on the (Hidden Markov Model) method or the like is provided.

As means for achieving a high recognition rate by a plurality of adaptive noise canceling sections and a speech recognition processing section, each of the outputs of each adaptive noise canceling section performs speech recognition. , The recognition result having the highest probability is selected from among the recognition results.

Further, as a means for achieving a high recognition rate by a plurality of adaptive noise canceling units and a speech recognizing unit working together, each output signal of each adaptive noise canceling unit is determined by a heuristic method (by a simple method). ), Signal to noise ratio (S / N
Ratio), an adaptive noise canceling unit having the largest S / N ratio is selected, speech recognition is performed on an output signal of the adaptive noise canceling unit, and the result is used as a recognition result.

Furthermore, in a heuristic method for the output signal of the adaptive noise canceling unit, the output signal is temporally divided into a speech section and a non-speech section as a method of obtaining an S / N ratio. The average value of the amplitude is obtained, and the ratio between the average AS of the amplitude in the vocal section obtained in this way and the average AN of the amplitude in the non-vocal section, and the S / N ratio for obtaining AS / AN are configured. .

FIG. 1 is a functional block diagram of a speech recognition system 60 installed in a car. In this FIG.
The voice recognition system 60 captures voice to be recognized, converts the captured voice signal into a digital signal 101, and supplies the digital signal 101 to the adaptive noise cancellation circuits 31 to 33.
10 and a circuit for capturing a specific noise signal used to remove a noise signal mixed in the captured audio signal 101, capturing a car audio signal in a car,
Adaptive noise cancellation circuit by converting to digital signal 211
A noise signal input circuit 21 provided to the adaptive noise canceling section 32; a noise signal input circuit 21 provided to the adaptive noise canceling section 32; A noise signal input circuit 33 for capturing an engine sound of a vehicle, converting the captured engine sound signal into a digital signal 231 and supplying the digital signal 231 to the adaptive noise canceling circuit 33.

Further, the voice recognition system 60 determines whether the voice signal is a voiced section or a silent section by judging from the captured voice signal 101 supplied from the voice signal input circuit 10,
A sound / silence detection control circuit 30 for providing a coefficient update command signal 301 for updating the filter coefficients (weighting coefficients) of the adaptive noise canceling circuits 31, 32, 33 in a silent section in response to this detection, and a digital filter. Using an adaptive noise canceling circuit 31 for removing the digital signal 211 obtained by capturing the car audio signal from the captured audio signal 101 and providing the removed audio signal 311 to the voice recognition circuit 40, using a digital filter, An adaptive noise canceling circuit 32 that removes the captured digital signal 221 of the noise signal outside the vehicle from the captured audio signal 101 and supplies the removed voice signal 321 to the voice recognition circuit 40, and a captured voice signal using a digital filter. The digital signal 231 of the captured engine sound signal is removed from the signal 101, and the removed audio signal is removed.
And an adaptive noise canceling circuit 33 for providing a speech recognition circuit 331 to the speech recognition circuit 40.

Further, the speech recognition system 60 separately performs speech recognition based on the Heiden-Markov model method for the removed speech signal 311, the removed speech signal 321, and the removed speech signal 331, and determines the probability with certainty. The speech recognition probability 41 for the removed speech signal 311 is given to the speech recognition probability comparison circuit 50, and the speech recognition probability 42 for the removed speech signal 321 is given to the speech recognition probability comparison circuit 50. A speech recognition circuit 40 for giving a speech recognition probability 43 to the speech signal 331 to the speech recognition probability comparison circuit 50, and a speech word having the highest speech recognition probability from the speech recognition probabilities 41, 42 and 43. And a speech recognition probability comparison circuit 50 which outputs the result as a recognition result 51.

The noise signal input circuit 21 directly takes in the line output electric signal as the sound output from the car audio system, and converts it into a digital signal 211 obtained by capturing the car audio signal.
May be configured to be supplied.

FIG. 2 is a diagram for explaining the operation of the speech recognition system 60. In this speech recognition system,
First, in the adaptive noise canceling circuit 31, the digital signal 211 obtained by capturing the car audio signal by the noise signal input circuit 21 is converted from the audio signal 101 captured by the audio signal input circuit 10.
Is removed (step S10), and the removed speech signal 311 is speech-recognized by the speech recognition circuit 40 by the Hidden Markov Model method (step S40), and the speech recognition symbol (speech recognition word) S1 and the speech recognition probability P1 Desired.

Also in the adaptive noise canceling circuit 32, the audio signal 101 captured by the audio signal input circuit 10
The digital signal 221 of the noise signal outside the vehicle captured by the noise signal input circuit 22 is removed (step S20), and the removed voice signal 321 is voice-recognized in the voice recognition circuit 40 by the Hidden Markov Model method (step S50). A recognition symbol S2 and a speech recognition probability P2 are obtained. Furthermore, in the adaptive noise canceling circuit 33, the audio signal input circuit
The digital signal 231 of the engine sound signal captured by the noise signal input circuit 23 is removed from the voice signal 101 captured in step 10 (step S30), and the removed voice signal 331 is converted to a voice recognition circuit.
At 40, speech recognition is performed by the Hidden Markov Model method (step S60), and a speech recognition symbol S3 and a speech recognition probability P3
Is required.

Next, in the speech recognition probability comparison circuit 50,
From these speech recognition probabilities P1, P2, P3, the speech recognition symbols S1, S2, S3 having the highest probability are selected and output. In the case of a car navigation system, the selected voice recognition symbol is used as a command for controlling the system.

In order to execute the above-described operation of the speech recognition system 60 shown in FIG.
33, the speech recognition circuit 40, and the speech recognition probability comparison circuit 50 can be performed by program processing. In order to perform such processing by program processing and to incorporate it in a car navigation system in a small size, it is preferable to mount the program in flash memory or ROM. Alternatively, the information may be stored in a magnetic disk device for backup.

FIG. 3 shows adaptive noise canceling circuits 31, 3
35 is a diagram of a configuration example of an example of 2, 33. FIG. The adaptive noise cancellation circuits 31, 32, and 33 described above can be realized with the same circuit configuration. In FIG. 3, adaptive noise canceling circuits 31, 32, and 33 are adaptive digital filter circuits.
312 and a subtraction circuit 313. The adaptive digital filter circuit 312 includes a digital signal 211 that captures a car audio signal in a vehicle, a digital signal 221 of a noise signal outside the vehicle, and a digital signal 231 of an engine sound signal.
Is given, a pseudo noise 3121 is generated to remove noise contained in the audio signal 101 captured by the audio signal input circuit 10 and is supplied to the subtraction circuit 313.

A subtraction circuit 313 converts an audio signal 101 captured by the audio signal input circuit 10 into an adaptive digital filter circuit 312.
The pseudo noise 3121 obtained in the above is subtracted to output the audio signal 311 from which the car audio signal has been removed, or the audio signal 321 from which the digital signal 221 of the noise signal outside the car has been removed.
Or an audio signal 331 from which the digital signal 231 of the engine sound signal has been removed. These audio signals
311, the audio signal 321, and the audio signal 331 are fed back to the adaptive digital filter circuit 312, and the filter coefficients are updated so that the residual component of the noise signal can be further reduced.

FIG. 4 shows the adaptive noise canceling circuit of FIG.
31 is a diagram illustrating a configuration example of an adaptive digital filter circuit 312 as an example of 31, 32, and 33. FIG. In FIG. 4, adaptive noise canceling circuits 31, 32, and 33 are a digital signal 211 that captures a car audio signal in a vehicle, a digital signal 221 of a noise signal outside the vehicle, or a digital signal of an engine sound signal.
231 and other delay circuits 3122 to 3125, and audio signals
311, a filter coefficient update circuit 3130 for updating a filter coefficient from the audio signal 321, the audio signal 331, and the like, and multipliers 3126 to 3129 for multiplying the filter coefficient by the delay signal.
And an adder 3131 that adds the respective multiplication results.

When the coefficient update command signal 301 is given from the sound / non-sound detection control circuit 30, the filter coefficient update circuit 3130 outputs the audio signal from which the car audio signal has been removed.
311, coefficient is updated and multiplied so that noise signal components mixed in the audio signal 321 from which the digital signal 221 of the noise signal outside the vehicle is removed, the audio signal 331 from which the digital signal 231 of the engine sound signal is removed, and the like are reduced. Vessels 3126, 3127, 312
8, give to 3129. In the sound section, the coefficient update is stopped and the digital signal 211 capturing the car audio signal in the car or the digital signal 221 of the noise signal outside the car is stopped.
Alternatively, the digital signal 231 of the engine sound signal is passed through delay circuits 3122 to 3125, and a multiplier is applied to the delayed signal.
The filter coefficients are multiplied by 3126 to 3129, the respective multiplication results are added by an adder 3131, and the addition result 3121 is output as a pseudo noise signal.

Although FIG. 4 shows a non-recursive digital filter, a recursive digital filter may be employed. Further, it may be constituted by a multi-stage digital filter of a recursive type and a non-recursive type.

FIG. 5 is a diagram showing a configuration example of an example of the speech recognition circuit 40. In FIG. 5, the speech recognition circuit 40 is an LPC
It comprises an analysis circuit 401, an HMM speech dictionary circuit 402, and a Viterbi matching circuit 403.

In the LPC analysis circuit 401, the audio signal 311 from which the input car audio signal has been removed, the audio signal 321 from which the digital signal 221 of the noise signal outside the vehicle has been removed, and the audio signal from which the digital signal 231 of the engine sound signal has been removed. From 331 etc., the audio waveform is divided into short sections (frame and length are, for example,
(About 10 msec to 30 msec) and extract feature parameters for each frame.

In this speech analysis, an LPC (Linear Predidiode) which is widely used as an efficient method adapted to the characteristics of the speech is used.
ction Coding (linear prediction) analysis and LP from LPC coefficient
Calculate C cepstrum. This LPC cepstrum
It is the inverse Fourier transform of the logarithmic spectrum, has properties similar to human auditory characteristics, and expresses speech efficiently with a relatively small number of parameters. In addition, a delta cepstrum, which is a temporal change amount of a cepstrum representing a dynamic characteristic of a spectrum, a logarithmic power representing a voice intensity, and a delta logarithmic power, which is a temporal change amount, are used as the feature parameters. The information of the LPC analysis result is provided to the Viterbi matching circuit 403.

The Viterbi matching circuit 403 is an HMM speech dictionary circuit.
Using 402, the HMM model representing phonemes and words is compared with the unknown input speech by the Viterbi algorithm to determine the similarity. That is, the speech recognition probability P is generated from the time series C of the speech feature vector from each word model M, and the word corresponding to the model having the maximum recognition probability is output as the speech recognition result.

In the above-described embodiment, an example has been described in which speech recognition is performed by the Hidden Markov Model method. Alternatively, speech recognition may be performed by a dynamic programming method.

Here, a simple example in which two noise signals, a car audio signal and a noise signal from outside the car, are canceled as noise signals will be described. In such a case, there are two factors that hinder recognition of the audio signal from the microphone: a noise signal from the car audio system and a noise signal from outside the vehicle. this house,
For example, a case where the noise signal from the car audio system is large and the noise signal from outside the car is small will be described.

In the adaptive noise canceling circuit 31 for removing the noise signal of the car audio system, the effect of the adaptive noise cancellation is exerted, and the noise signal by the car audio system during the microphone phone input can be reduced. An audio signal close to the audio signal can be output. As a result, the speech recognition circuit 40 given this speech signal can recognize a correct speech signal as the word S1, and at the same time, can obtain a high value of the certainty P1 (speech recognition rate).

On the other hand, in the adaptive noise canceling circuit 32 for removing a noise signal from outside the car, the noise component outside the car, which is a small part of the noise component, can be reduced to some extent, but the car audio, which is a larger noise component, can be reduced. The noise signal from the system is not reduced. Therefore,
The noise signal from the car audio system remains mixed in the output signal of the adaptive noise cancellation circuit 32. In the speech recognition circuit 40, since the noise signal from the car audio system is given while being mixed, an erroneous recognition word S2 is easily output, and the speech recognition probability P2 also has a low value.

From the above, in the speech recognition probability comparison circuit 50, a speech recognition probability higher than the speech recognition probability P2 is obtained.
The recognition word S1 of P1 is selected and output.

Conversely, when the noise signal outside the car is large and the noise signal from the car audio system is small, the recognition word of P2 having a higher speech recognition probability than the speech recognition probability P1 is obtained.
S2 is selected and output. In this way, by selecting the recognition result of the output of the adaptive noise cancellation circuit that maximizes the likelihood, a noise signal that has the worst effect on speech recognition among a plurality of noise sources is identified. By recognizing the audio signal from which the noise signal that has the worst effect has been removed, the most reliable recognition result can be obtained.

As described above, a plurality of noise sources are individually subjected to noise elimination using an adaptive noise canceling circuit.
Among them, the most probable speech recognition word of the speech recognition result is selected, and a noise signal that has the least adverse effect on speech recognition among a plurality of noise sources is identified and the bad signal is given. Since the configuration is such that the noise signal is removed, it is possible to improve the speech recognition accuracy in a noisy acoustic environment. Moreover, since the calculation processing is not complicated and there is no element that increases the processing amount, it can be easily realized.

The speech recognition system 60 of the first embodiment described above.
In the above, speech recognition can be performed with a simple configuration for noise signals, but speech recognition is performed for all outputs of the adaptive noise cancellation circuits 31, 32, and 33 corresponding to each noise signal. As the number of noise signals generated increases, the amount of calculation may increase. Therefore, in the second embodiment, before performing speech recognition, it is checked which output signal of the adaptive noise canceling circuit is a signal of good quality, and the output signal of the adaptive noise canceling circuit of good quality is checked. It is configured to perform voice recognition only when Thus, a good speech recognition system can be realized with a smaller amount of calculation.

FIG. 6 shows a speech recognition system according to the second embodiment.
FIG. 70 is a functional configuration diagram of 70. In FIG. 6, a speech recognition system 70 includes a speech signal input circuit 10 and a noise signal input circuit.
A voice / silence section is detected from the voice signal 101 from the voice signal input circuit 10 and a coefficient update command signal 301 is output and supplied to the adaptive noise cancellation circuits 31 to 33, and voice / silence section detection is performed. A sound / silence detection control circuit 30 for providing a signal 302 to an adaptive noise canceling signal selecting circuit 80, and output signals 311, 321, 3 of adaptive noise canceling circuits 31 to 33
An adaptive noise canceling signal selection circuit 80 for determining the S / N ratio of each of the 31 and selecting the output signal having the highest S / N ratio
And a speech recognition circuit 90 for performing speech recognition for an adaptive noise canceling output signal having a high S / N ratio by a Heiden-Markov model method or a dynamic programming method and outputting a recognition result 901.
It is composed of

FIG. 7 is a diagram for explaining the operation of the speech recognition system according to the second embodiment of FIG. In FIG. 7, first, the adaptive noise canceling circuit 31 removes the digital signal 211 from which the car audio signal was captured by the noise signal input circuit 21 (step S10), and the removed audio signal 311 is used as an adaptive noise canceling signal selecting circuit. Given to 80.

In the adaptive noise canceling circuit 32, the audio signal 101 captured by the audio signal input circuit 10
The digital signal 221 of the noise signal outside the vehicle captured by the noise signal input circuit 22 is removed (step S20), and the removed voice signal 321 is supplied to the adaptive noise canceling signal selection circuit 80. Further, also in the adaptive noise canceling circuit 33, from the audio signal 101 captured by the audio signal input circuit 10,
The digital signal 231 of the engine sound signal captured by the noise signal input circuit 23 is removed (step S30), and the removed audio signal 331 is supplied to the adaptive noise canceling signal selection circuit 80.

In the adaptive noise canceling signal selecting circuit 80, the S / N of the output signal 311 of the adaptive noise canceling circuit 31
The ratio is obtained and this value is set as S / N1 (step S80). Further, the S / N of the output signal 321 of the adaptive noise cancel circuit 32
The ratio is obtained and this value is set as S / N2 (step S90). Further, the S / N of the output signal 331 of the adaptive noise canceling circuit 33 is
The ratio is obtained and this value is set as S / N3 (step S100). Select the output signal 801 having the largest S / N ratio (step
(S110), and outputs the speech recognition result 901 to the speech recognition circuit 90 (step S120).

In order to realistically execute the operation of the speech recognition system 70 shown in FIG.
33, the adaptive noise canceling signal selection circuit 80, the speech recognition circuit 90, and the like can be performed by program processing. It is executed by the program processing that performs such processing,
Moreover, in order to incorporate the program in a car navigation system in a small size, it is better to load the program into flash memory or ROM. Alternatively, the information may be stored in a magnetic disk device for backup.

FIG. 8 is a diagram for explaining how to obtain the above-mentioned S / N ratio. In FIG. 8, the output signals 331, 321 and 331 of the adaptive noise cancellation circuits 31 to 33 are
Based on the sound / silence section detection signal 302 from the sound / silence detection control circuit 30, the sound section (speech section 82) where the user is speaking is determined, and other than that, that is, the user is speaking. Silence section judged not to exist (non-speech section 81)
The average level AN over the signal amplitude section in the non-speech section 81 and the average level AS over the signal amplitude section in the utterance section 82 are obtained, and the ratio AN / AS is calculated as S / N It is good to be a ratio.

As described above, according to the configuration of the speech recognition system of the second embodiment, adaptive noise cancellation is performed on each noise signal, and the S / N ratio of each adaptive noise cancellation output signal is obtained. Since speech recognition is performed on the output signal with the highest S / N ratio among these, the speech recognition accuracy can be improved even in a noisy environment while reducing the amount of calculation. become. In addition,
In order to determine the quality of the adaptive noise canceling output signal, the amount of signal distortion may be measured, and the signal quality may be determined based on the magnitude of the amount of signal distortion (eg, the amount of harmonic distortion).

[0049]

As described above, according to the present invention, when noise signals from a plurality of noise sources are mixed in a main audio signal, a noise signal from a certain noise source is removed from the main audio signal. Speech recognition of the output main acoustic signal of one adaptive noise canceling means for performing a speech recognition using a statistical acoustic model, obtaining a recognition word and obtaining a certainty of the recognition, and generating a noise signal from another noise source. The output main acoustic signal of another adaptive noise canceling means for removing is also speech-recognized using the statistical acoustic model, and a recognition word is obtained and its recognition certainty is obtained.From these recognition results,
By configuring so that a recognition word having a high probability of speech recognition is output as a recognition result, for example, a sound in which a plurality of different types of noise signals exist, such as noise from the outside of a car or a sound signal from a car audio system. In an environment, the recognition accuracy of speech recognition can be improved with a small amount of processing.

[Brief description of the drawings]

FIG. 1 is a functional configuration diagram of a voice recognition system installed in an automobile in a first embodiment of the present invention.

FIG. 2 is a diagram for explaining the operation of the speech recognition system of the embodiment shown in FIG.

FIG. 3 is a functional configuration diagram of an adaptive noise canceling circuit of the speech recognition system of the embodiment.

4 is a functional configuration diagram of an adaptive digital filter circuit of the adaptive noise canceling circuit of the embodiment shown in FIG.

5 is a functional configuration diagram of a speech recognition circuit of the speech recognition system of the embodiment shown in FIG.

FIG. 6 is a functional configuration diagram of a speech recognition system according to a second embodiment.

FIG. 7 is a diagram for explaining the operation of the speech recognition system according to the embodiment shown in FIG. 6;

FIG. 8 is an explanatory diagram of how to calculate an S / N ratio in an adaptive noise canceling signal selection circuit of the speech recognition system shown in FIG.

[Explanation of symbols]

 10 Speech signal input circuit 21 ~ 23 Noise signal input circuit 31 ~ 33 Adaptive noise cancellation circuit 40 Speech recognition circuit 50 Speech recognition probability comparison circuit

Claims (9)

[Claims] 1. A speech recognition system comprising a main sound signal capturing means capable of capturing main sound and outputting a main sound signal and capturing noise from at least two noise sources, wherein the system comprises at least two or more sound sources. A system comprising at least two noise signal capturing means and adaptive noise canceling means for removing a noise signal from a noise source, wherein the system captures noise from a noise source and performs noise capture. A first noise signal capturing unit that outputs a signal, a second noise signal capturing unit that captures noise from another noise source and outputs a noise capturing signal, and the first noise signal capturing from the main acoustic signal. First noise canceling means for removing the noise capture signal captured by the means and outputting the removed main audio signal; and means for capturing the second noise signal from the main audio signal. Second adaptive noise canceling means for removing the noise capture signal captured in the stage and outputting the removed main acoustic signal; and using a statistical acoustic model for a main acoustic signal output from the first adaptive noise canceling means. And the likelihood of the recognition is determined and the likelihood of the recognition is determined. The output main acoustic signal of the second adaptive noise canceling means is also subjected to voice recognition using the statistical acoustic model to determine the recognition word. And a voice recognition means for obtaining a certainty of the recognition and outputting a recognized word having a high certainty of the voice recognition as a recognition result from the recognition results. 2. The speech recognition system according to claim 1, wherein the speech recognition is performed using a Heiden-Markov model method as a statistical acoustic model of the speech recognition. . 3. A speech recognition system including main sound signal capturing means capable of capturing main sound and outputting a main sound signal and capturing noise from at least two noise sources, wherein the system comprises at least two or more sound sources. A system comprising at least two noise signal capturing means and adaptive noise canceling means for removing a noise signal from a noise source, wherein the system captures noise from a noise source and performs noise capture. A first noise signal capturing unit that outputs a signal, a second noise signal capturing unit that captures noise from another noise source and outputs a noise capturing signal, and the first noise signal capturing from the main acoustic signal. First noise canceling means for removing the noise capture signal captured by the means and outputting the removed main audio signal; and means for capturing the second noise signal from the main audio signal. A second adaptive noise canceling means for removing the noise capture signal captured by the stage and outputting the removed main acoustic signal; and using a dynamic programming method for the main acoustic signal output from the first adaptive noise canceling means. And the similarity and the word thereof are obtained, and the main sound signal output from the second adaptive noise canceling means is also subjected to speech recognition using the dynamic programming method to obtain the similarity and the word. And a speech recognition unit that outputs a recognized word having a high degree of similarity from the recognition results as a recognition result. 4. A speech recognition system comprising main sound signal capturing means capable of capturing main sound and outputting a main sound signal and capturing noise from at least two noise sources, wherein the system comprises at least two or more sound sources. A system comprising at least two noise signal capturing means and adaptive noise canceling means for removing a noise signal from a noise source, wherein the system captures noise from a noise source and performs noise capture. A first noise signal capturing unit that outputs a signal, a second noise signal capturing unit that captures noise from another noise source and outputs a noise capturing signal, and the first noise signal capturing from the main acoustic signal. First noise canceling means for removing the noise capture signal captured by the means and outputting the removed main audio signal; and means for capturing the second noise signal from the main audio signal. A second adaptive noise canceling unit for removing the noise capture signal captured by the stage and outputting the removed main acoustic signal; a main acoustic signal output from the first adaptive noise canceling unit and the second adaptive noise Voice recognition means for determining the signal quality of the output main sound signal of the canceling means, selecting the output main sound signal having good signal quality, and performing voice recognition on the selected output main sound signal. Voice recognition system. 5. The speech recognition system according to claim 4, wherein the system determines the signal quality by determining a signal-to-noise ratio and / or a signal distortion amount of an output main audio signal. Voice recognition system. 6. The speech recognition system according to claim 4, wherein said speech recognition means performs said speech recognition by a statistical acoustic model method or a dynamic programming method. 7. A computer which captures main sound and outputs a main sound signal, and at least two or more main sound signals captured by main sound signal capturing means capable of capturing noise from at least two noise sources. A first noise signal acquisition unit that includes at least two or more noise signal acquisition units for acquiring a noise signal from a noise source, and acquires noise from a certain noise source and outputs a noise acquisition signal; and Recognition control for removing the noise capture signal captured by the second noise signal capture means that captures noise from the noise source and outputs the noise capture signal, and performs voice recognition on the removed main acoustic signal A recording medium on which a program is recorded, wherein the voice recognition control program is configured to remove at least two or more of the noise capture signals from the main audio signal. And two or more adaptive noise canceling steps, wherein the noise capturing signal captured by the first noise signal capturing means is removed from the main acoustic signal, and the removed main acoustic signal is output. 1 adaptive noise canceling step; and 2nd adaptive noise canceling step of removing the noise capture signal captured by the second noise signal capturing means from the main audio signal and outputting the removed main audio signal. The main acoustic signal output from the first adaptive noise canceling step is subjected to speech recognition using a statistical acoustic model to determine a recognition word and the likelihood of the recognition; An audio signal is also subjected to speech recognition by the statistical acoustic model, and a recognition word is determined and a probability of the recognition is determined.
A speech recognition step of outputting, as a recognition result, a recognition word having a high probability of speech recognition from the recognition results. 8. A computer that captures main sound and outputs a main sound signal, and at least two or more main sound signals captured by main sound signal capturing means capable of capturing noise from at least two noise sources. First noise signal capturing means for capturing noise from one noise source and outputting a noise capturing signal, and at least two noise signal capturing means for capturing a noise signal from the noise source of the other type. A speech recognition control program for removing a noise capture signal captured by a second noise signal capture unit that captures noise from a noise source and outputs a noise capture signal, and performs voice recognition on the removed main acoustic signal. Recording medium, wherein the voice recognition control program is configured to reduce at least two or more of the noise capture signals from the main acoustic signal. Also includes two or more adaptive noise canceling steps, wherein the noise capturing signal captured by the first noise signal capturing means is removed from the main audio signal, and the removed main acoustic signal is output. 1 adaptive noise canceling step; and 2nd adaptive noise canceling step of removing the noise capture signal captured by the second noise signal capturing means from the main audio signal and outputting the removed main audio signal. The output main audio signal of the first adaptive noise canceling step is subjected to speech recognition using dynamic programming to determine a similarity and its word, and the output main audio signal of the second adaptive noise canceling step is also obtained. A speech recognition step of performing speech recognition using the dynamic programming to obtain a similarity and a word thereof, and outputting a recognition word having a high similarity as a recognition result from the recognition results. And a recording medium storing a speech recognition control program. 9. A computer which captures main sound and outputs a main sound signal, and at least two or more main sound signals captured by main sound signal capturing means capable of capturing noise from at least two noise sources. A first noise signal acquisition unit that includes at least two or more noise signal acquisition units for acquiring a noise signal from a noise source, and acquires noise from a certain noise source and outputs a noise acquisition signal; and Recognition control for removing the noise capture signal captured by the second noise signal capture means that captures noise from the noise source and outputs the noise capture signal, and performs voice recognition on the removed main acoustic signal A recording medium on which a program is recorded, wherein the voice recognition control program is configured to remove at least two or more of the noise capture signals from the main audio signal. And two or more adaptive noise canceling steps, wherein the noise capturing signal captured by the first noise signal capturing means is removed from the main acoustic signal, and the removed main acoustic signal is output. 1 adaptive noise canceling step; and 2nd adaptive noise canceling step of removing the noise capture signal captured by the second noise signal capturing means from the main audio signal and outputting the removed main audio signal. Determining the signal quality of the output main audio signal of the first adaptive noise cancellation step and the signal quality of the output main audio signal of the second adaptive noise cancellation step, and selecting the output main audio signal having good signal quality; And a voice recognition step of performing voice recognition on the output main sound signal.