JP2003099100A - Speech recognition apparatus and method - Google Patents

Abstract

(57) Abstract: The present invention provides a speech recognition apparatus and method that can improve recognition performance even when a plurality of audio signal output units are provided. SOLUTION: The speech recognition device of the present invention includes an acoustic signal generating means 10 for generating an acoustic signal, a plurality of acoustic signal output means 21 and 22 for outputting the acoustic signal as sound waves, and an acoustic signal input means 30 for inputting voice. And the acoustic signal input means 30
An output signal of the audio signal generating means 10 and an output signal of the audio signal generating means 10, and an interference sound removing unit 100 for reducing output signal components of the plurality of audio signal output means 21 and 22 mixed in the output signal of the audio signal input means 30; Speech recognizing means 40 which receives an output signal of the interfering sound removing unit 100 and recognizes speech is provided.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition capable of reducing the output signal from the speaker or the like and performing the voice recognition with high performance when the voice recognition is performed in an environment where an acoustic signal is output from the speaker or the like. An apparatus and its method.

[0002]

2. Description of the Related Art As a conventional voice recognition device using an echo canceller, there is one described in, for example, Japanese Patent Laid-Open No. 5-323993.

In this conventional voice recognition device, when a voice response sound output by the voice interaction system itself is input to the voice input microphone, a process of canceling the voice response sound is performed, and the voice recognition is enabled even when the voice response is output. ing.

Further, for example, in the device described in Japanese Patent Laid-Open No. 7-175498, when the voice is recognized via the telephone line, the guidance voice sent to the user is reduced by using the echo canceller. Is configured,
The guidance voice is prevented from being mixed with the recognition target voice.

Furthermore, in the device disclosed in Japanese Patent Laid-Open No. 8-107375, the sound signal reproduced by the device itself is canceled in front of the sound recording unit for the purpose of reproducing the sound signal and simultaneously recording the sound. It has a simple structure.

[0006]

However, the voice recognition device having the conventional echo canceller means has a problem that it cannot cope with the case where there are a plurality of means for outputting an acoustic signal.

The present invention has been made to solve the conventional problems, and an object of the present invention is to provide a speech recognition apparatus which can improve the recognition performance even when it has a plurality of acoustic signal output means.

[0008]

A speech recognition apparatus of the present invention comprises an acoustic signal generating means for generating an acoustic signal, and a plurality of acoustic signal outputting means for outputting the acoustic signal as a sound wave.
A sound signal input means for inputting voice, an output signal of the sound signal input means and an output signal of the sound signal generation means are input, and the plurality of sound signal output means mixed in the output signal of the sound signal input means A configuration is provided that includes an interfering sound removing unit that reduces an output signal component and a voice recognizing unit that inputs the output signal of the interfering sound removing unit and recognizes the voice.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition and the quality of the voice is deteriorated, Can be eliminated and the voice recognition performance can be improved.

The acoustic signal generating means receives the output signal of a predetermined external acoustic signal generating means as an input and outputs the acoustic signal to the plurality of acoustic signal outputting means.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the quality of the voice, Can be eliminated and the voice recognition performance can be improved.

Further, the interfering sound removing section has a plurality of stages of echo canceller means for reducing the output signal component of the acoustic signal output means mixed in the output signal of the acoustic signal input means, The echo canceller means inputs the output signal of the acoustic signal input means, the echo canceller means of the other stage inputs the output signal of the echo canceller means of the preceding stage, and each echo canceller means has the acoustic signal generating means. One of the output signals is input, and the output signal component of the acoustic signal output unit mixed in the output signal of the acoustic signal input unit is reduced and output.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the quality of the voice, The voice signal can be removed and the voice recognition performance can be improved.

The disturbing sound removing section detects a plurality of levels of the echo canceller means and a plurality of signal levels of the output signals of the acoustic signal generating means inputted to the plurality of echo canceller means, respectively. The detection means and the output signals of the plurality of signal level detection means are input, and it is determined from the signal levels which one of the plurality of stages of echo canceller means is to perform the processing for improving the performance of the echo canceller. However, the learning timing determining means determines the learning timing for improving the echo canceller performance.

With this configuration, when recognizing the voice generated by the speaker, the performance of the process of removing the interfering sound can be improved, and the acoustic signals output from the plurality of speakers are input microphones for voice recognition. It is possible to reliably remove the voice signal even when the quality of the voice is deteriorated by being mixed in the voice recognition performance and improve the voice recognition performance.

The interfering sound removing section inputs any one of the plurality of output signals from the acoustic signal generating means and the output signal of the acoustic signal input means, and mixes them with the output signal of the acoustic signal input means. A plurality of filter coefficient calculation means for calculating the filter coefficient of the filter that acts to attenuate one output signal of the acoustic signal output means,
A plurality of signal level detecting means for detecting the level of each output signal from the acoustic signal generating means input to each of the plurality of filter coefficient calculating means, and inputting the output signals of the plurality of signal level detecting means, A learning timing determination unit that determines which of the filter coefficient calculation units should learn the filter coefficient from the signal level and determines the learning timing, and the plurality of filter coefficient calculation units. Filter coefficient integrating means for integrating the filter coefficients calculated in
And a filtering unit for inputting the output signal of the filter coefficient integrating unit and the audio signal input unit, and for performing a process of reducing the audio signal output component after using the filter coefficient obtained by the filter coefficient integrating unit. And

With this configuration, when recognizing the voice generated by the speaker, the performance of the process of removing the interfering sound can be improved, and the efficient calculation can be performed.
As a result, even when the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition and the quality of the voice is deteriorated, the voice signal can be reliably removed and the voice recognition performance can be improved. .

In the learning timing determining means, all the output signals except one of the output signals of the plurality of signal level detecting means are less than the threshold value, and the remaining one output signal is less than the threshold value. In addition, when it is equal to or more than a threshold value set separately, learning of the echo canceller means or the filter coefficient calculation means that receives the acoustic signal input to the signal level detection means is performed.

With this configuration, it is possible to improve the performance of the process of removing the interfering sound when the voice generated by the speaker is recognized, and the acoustic signals output from the plurality of speakers are input microphones for voice recognition. It is possible to reliably remove the voice signal even when the quality of the voice is deteriorated by being mixed in the voice recognition performance and improve the voice recognition performance.

Further, the apparatus further comprises a noise suppressing means for suppressing noise by inputting the output signal of the acoustic signal inputting means, and the interfering sound removing section outputs the output signal of the noise suppressing means and the acoustic signal generating means. The output signal is input and the output signal component of the acoustic signal output means mixed in the output signal of the acoustic signal input means is reduced.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition and the voice quality is deteriorated, the voice signal is also deteriorated. Is eliminated and noise is suppressed, the voice recognition performance can be further improved.

Further, the speech recognition means further comprises noise suppressing means for suppressing noise by inputting the output signal of the interfering sound removing portion, and the voice recognizing means inputs the output signal of the noise suppressing means and the speaker utters. I decided to recognize the voice.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the quality of the voice, Is eliminated and noise is suppressed, the voice recognition performance can be further improved.

Further, first noise suppressing means for suppressing noise by inputting the output signal of the acoustic signal inputting means,
A second noise suppressing unit for inputting an output signal of the interfering sound removing unit to suppress noise, and the interfering sound removing unit includes an output signal of the first noise suppressing unit and the acoustic signal generating unit. Output signal of the acoustic signal output means mixed in the output signal of the first noise suppression means is reduced, and the voice recognition means outputs the output signal of the second noise suppression means. We decided to recognize the voice generated by the speaker by inputting a signal.

With this configuration, when the voice generated by the speaker is recognized, even if the sound quality output from a plurality of speakers is mixed into the input microphone for voice recognition to deteriorate the voice quality, Is eliminated and noise is suppressed, the voice recognition performance can be further improved.

Further, there is further provided acoustic signal output control means for controlling the output of the acoustic signal generation means, wherein the acoustic signal output control means receives the output signal of the acoustic signal generation means and receives the plurality of acoustic signals. It is decided to control the signal output to the output unit and the disturbing sound removing unit.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the quality of the voice, Can be eliminated and the voice recognition performance can be improved.

Further, it further comprises a transfer characteristic measuring signal generating means for generating a signal suitable for measuring the transfer characteristic, and the acoustic signal output control means and the output signal of the acoustic signal generating means and the The transfer characteristic measuring signal generating means is controlled so as to output one of the output signals.

With this configuration, when recognizing the voice generated by the speaker, the performance of the process of removing the interfering sound can be further improved, and the acoustic signals output from the plurality of speakers are used for the voice recognition input. It is possible to improve the voice recognition performance by removing the voice signal even when the voice quality is deteriorated by being mixed in the microphone.

The signal output from the transfer characteristic measuring signal generating means is either white noise or an M-sequence signal.

With this configuration, when recognizing the voice generated by the speaker, the performance of the process of removing the interfering sound can be further improved, and the acoustic signals output from the plurality of speakers are used for the voice recognition input. It is possible to improve the voice recognition performance by removing the voice signal even when the voice quality is deteriorated by being mixed in the microphone.

Further, the apparatus further comprises noise suppressing means for suppressing noise from the information about the output signal of the acoustic signal inputting means and the signal output from the acoustic signal output controlling means, and the interfering sound removing section includes the noise suppressing means. The output signal of the suppressing means is input and processed.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the voice quality, Is eliminated and noise is suppressed, the voice recognition performance can be further improved.

Further, the apparatus further comprises noise suppressing means for suppressing noise from the information on the output signal of the interfering sound removing section and the signal output from the acoustic signal output control means,
The voice recognition means inputs the output signal of the noise suppression means and recognizes the voice uttered by the speaker.

With this configuration, when a speaker recognizes a voice generated, even if the sound signal output from a plurality of speakers is mixed in the input microphone for voice recognition and the voice quality is deteriorated, the voice signal is also deteriorated. Is eliminated and noise is suppressed, the voice recognition performance can be further improved.

Further, the system further comprises a synthetic sound generating means for creating a synthetic sound to be heard by a speaker, and a voice dialogue control means for controlling the voice recognition result to realize a voice dialogue. The signal output control means inputs the output signal of the acoustic signal generation means, the output signal of the synthesized sound generation means, and the output signal of the voice interaction control means, and outputs the output signal of the acoustic signal generation means and the One of the output signals of the synthetic sound generation means is output.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the voice quality, Can be improved to improve the voice recognition performance, and by realizing the dialogue with the speaker, it becomes possible for the speaker to easily recognize the voice.

Further, a display means for displaying the contents determined by the voice dialogue control means is further provided.

With this configuration, the dialogue with the speaker becomes easy for the user to understand.

The signal output from the acoustic signal generating means is a stereo signal.

With this configuration, when the voice generated by the speaker is recognized, even if the sound signal output from the stereo speaker is mixed in the input microphone for voice recognition and the voice quality is deteriorated, the voice signal is also output. It can be removed to improve the voice recognition performance.

Further, a plurality of signals output from the acoustic signal generating means have the same content.

With this configuration, even when the voice generated by the speaker is recognized, even if the acoustic signal output from the plurality of speakers is mixed with the input microphone for voice recognition and the quality of the voice is deteriorated. The voice signal can be removed and the voice recognition performance can be improved.

Furthermore, the voice recognition method of the present invention comprises: an acoustic signal generation procedure for generating an acoustic signal; a plurality of acoustic signal output procedures for outputting the acoustic signals as sound waves; and an acoustic signal input procedure for inputting voice. Input the output signal in the acoustic signal input procedure and the output signal in the acoustic signal generation procedure,
An interference sound elimination procedure for reducing output signal components in the plurality of acoustic signal output procedures mixed in the output signal in the acoustic signal input procedure, and inputting an output signal of the interference sound elimination procedure,
The voice recognition procedure for recognizing the voice is performed.

According to this method, when the voice generated by the speaker is recognized, even if the sound signals output from the plurality of speakers are mixed in the input microphone for voice recognition to deteriorate the voice quality, Can be eliminated and the voice recognition performance can be improved.

[0046]

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

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

As shown in FIG. 1, the speech recognition apparatus of the first embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10 are input, and are mixed in the output signal of the acoustic signal input means 30 with the first acoustic signal output means 21 and the second acoustic signal. Output means 22
And an audio signal recognizing means 40 which receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker.

The operation of the speech recognition apparatus of the first embodiment configured as described above will be described.

First, when the acoustic signal generating means 10 generates, for example, music, and the first acoustic signal outputting means 21 and the second acoustic signal outputting means 22 are outputting the sound waves to the space, the speaker speaks. A case of performing voice recognition will be described as an example.

The voice signal uttered by the speaker is passed to the interfering sound removing section 100 through the acoustic signal input means 30. At this time, since the signal input to the interfering sound removal unit 100 also includes the acoustic signals (for example, music) output from the first acoustic signal output means 21 and the second acoustic signal output means 22, If the voice is recognized as it is, a good recognition result cannot be obtained.

Therefore, the interfering sound removing section 100 performs a process of reducing the acoustic signals (for example, music) output from the acoustic signal output means 21 and 22 by utilizing the outputs of the acoustic signal input means 30 and the acoustic signal generation means 10. To do. The voice processed by the interfering sound remover 100 is passed to the voice recognition means 40,
Performs voice recognition processing.

As described above, according to the first embodiment, the first
In the voice recognition device having the acoustic signal output unit 21 and the second acoustic signal output unit 22 of FIG.
By providing the sound signal, the sound signals output from the first sound signal output means 21 and the second sound signal output means 22 are mixed into the sound signal input means 30 when the speaker utters the sound. The signal component present can be reduced, and the voice recognition performance can be improved.

In the first embodiment of the present invention, the first
The case where the sound signal output means 21 and the second sound signal output means 22 are provided is taken as an example, but the sound signal output means 3
Similar processing can be performed when there are two or more.

FIG. 2 shows a block diagram of a speech recognition apparatus according to the second embodiment of the present invention.

As shown in FIG. 2, the speech recognition apparatus according to the second embodiment has an external acoustic signal generating means 200 and an acoustic signal for outputting an acoustic signal by inputting an output signal of the external acoustic signal generating means 200. The signal generating means 10, the first acoustic signal outputting means 21 and the second acoustic signal outputting means 22 for outputting the output signal of the acoustic signal generating means 10 to the space as a sound wave.
, The acoustic signal input means 30 for inputting the voice uttered by the speaker, the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10, and the acoustic signal input means 30.
First acoustic signal output means 21 mixed in the output signal of
And the interfering sound removing unit 100 that reduces the output signal component of the second acoustic signal outputting unit 22, and the voice recognizing unit 40 that receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker. It is configured.

The operation of the speech recognition apparatus of the second embodiment configured as described above will be described. However,
The second embodiment of the present invention differs from the first embodiment only in the operations of the acoustic signal generating means 10 and the external acoustic signal generating means 200, and therefore only these two parts will be described.

First, an audio device such as a CD or a radio is connected to the acoustic signal generating means 10 as an external acoustic signal generating means 200. The acoustic signal generating means 10 receives the output signal of the external acoustic signal generating means 200 through a connection terminal or the like. Furthermore, in the acoustic signal generating means 10,
The input acoustic signal is passed to the first acoustic signal output means 21 and the second acoustic signal output means 22 for the acoustic signal, and is output to the space as a sound wave. At this time,
The acoustic signal generating means 10 can change the amplitude magnification of the output of the external acoustic signal generating means 200, which is an input signal, or change the number of channels of the output signal.

According to the second embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided. By connecting to the external acoustic signal generating means 200, the first acoustic signal outputting means 21 and the second acoustic signal outputting means 21 can be used when recognizing the voice generated by the speaker.
The signal component of the acoustic signal output from the acoustic signal output unit 22 mixed in the acoustic signal input unit 30 can be reduced, and the voice recognition performance can be improved.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 3 shows a block diagram of a speech recognition apparatus according to the third embodiment of the present invention.

As shown in FIG. 3, the speech recognition apparatus according to the third embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10 are input, and are mixed in the output signal of the acoustic signal input means 30 with the first acoustic signal output means 21 and the second acoustic signal. Output means 22
And an audio signal recognizing means 40 which receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker.

Further, the disturbing sound removing section 100 reduces the output signal components of the first acoustic signal output means 21 and the second acoustic signal output means 22 mixed in the output signal of the acoustic signal input means 30. One-stage echo canceller means 11
It is composed of first and second-stage echo canceller means 112.

The operation of the speech recognition apparatus of the third embodiment configured as described above will be described. However,
Since the configuration other than the interference sound removal unit 100 is the same as that of the first embodiment, only the internal operation of the interference sound removal unit 100 will be described.

From the acoustic signal input means 30 to the interfering sound removing section 1
The acoustic signal input to 00 is echo canceller means 111 of the first stage and echo canceller means 112 of the second stage.
Are successively processed by the voice recognition means 40 and passed to the voice recognition means 40.

First, the echo canceller means 111 of the first stage removes the signal output from the first acoustic output means 21 and mixed in the acoustic signal input means 30, and the echo canceller means 112 of the second stage. Passed.

Next, the second-stage echo canceller means 112
Then, the signal outputted from the second sound output means 22 and mixed in the sound signal input means 30 is removed and outputted.

Here, the first-stage echo canceller means 1
In order to describe a specific example of the processing performed by the 11th and second-stage echo canceller means 112, the case of using a typical echo canceller will be described with reference to the block diagram of the echo canceller in FIG.

The echo canceller is described in the document "Acoustic System and Digital Processing" (Oga, Yamazaki, Kaneda, published by The Institute of Electronics, Information and Communication Engineers).

First, consider a state in which no voice is uttered in FIG. At this time, when the time series signal of the acoustic signal radiated to the space from the acoustic signal output means 20 is x (k), x (k) is radiated to the space and the spatial transfer characteristic h '(k) is convolved. The signal y '(k) is input to the voice signal input means 30. At this time, the adaptive filter means 170 uses the acoustic signal x (k) and the output y ′ (k) of the acoustic signal input means 30 to determine the spatial transfer characteristic h.
(k) is estimated, the transfer characteristic h (k) is convolved with x (k), the input signal from the acoustic signal input means 30 is estimated, and y (k) is output. The y (k) is y'output from the acoustic signal input means 30.
By subtracting from (k), the acoustic echo (the signal output from the acoustic signal output means 20 entering the acoustic signal input means 30) can be removed.

Next, as described above, the transfer characteristic of the space is set to h (k)
If the signal z (k) uttered by the speaker is input with the filter being stable, the acoustic signal entering the voice signal input means 30 is s (k) = y ′ (k) + z (k However, the acoustic echo can be removed by the processing of s (k) = y '(k) + z (k) -y (k) in the echo canceller means 110.

This is the basic principle of the echo canceller, and the first-stage echo canceller means 111 and the second
The echo canceller means 112 of the stage performs such processing.

According to the third embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the echo canceller means of the first stage is used. By providing the interfering sound removing unit 100 having the 111 and the second-stage echo canceller means 112, the first acoustic signal output means 21 and the second acoustic signal output means when the voice generated by the speaker is recognized. Two
It is possible to reduce the signal component in which the acoustic signal output from No. 2 is mixed in the acoustic signal input means 30 and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

In the third embodiment of the present invention, the first
The case of having the echo canceller means 111 of the second stage and the echo canceller means 112 of the second stage is taken as an example, but the same processing can be performed when there are three or more echo canceller means.

Further, there are a plurality of echo canceller algorithms, and processing may be performed by a method other than the algorithm described here.

FIG. 4 shows a block diagram of a speech recognition apparatus according to the fourth embodiment of the present invention.

As shown in FIG. 4, the speech recognition apparatus of the fourth embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10 are input, and are mixed in the output signal of the acoustic signal input means 30 with the first acoustic signal output means 21 and the second acoustic signal. Output means 22
And an audio signal recognition unit 40 that receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker.

Further, in the disturbing sound removing section 100, the first stage for reducing the output signal components of the first acoustic signal output means 21 and the second acoustic signal output means 22 mixed from the output signal of the acoustic signal input means 30. Echo canceller means 11
First and second stage echo canceller means 112;
Acoustic signal generating means 1 input to the echo canceller means 111 in the second stage and the echo canceller means 112 in the second stage
First signal level detecting means 121 and second signal level detecting means 122 for detecting the level of the output signal of 0;
The output signals of the first signal level detecting means 121 and the second signal level detecting means 122 are input, and the echo canceller of the first-stage echo canceller means 111 and the second-stage echo canceller means 112 of the echo canceller is inputted from the signal levels. The learning timing determining means 130 is configured to determine which is the one to perform the processing for improving the performance, and to determine the learning timing for improving the echo canceller performance.

The operation of the speech recognition apparatus of the fourth embodiment constructed as above will be described. However,
Since the configuration other than the interference sound removal unit 100 is the same as that of the third embodiment, only the internal operation of the interference sound removal unit 100 will be described.

The acoustic signal input from the acoustic signal 30 to the interfering sound removing section 100 is the echo canceller means 1 of the first stage.
11 and the echo canceller means 112 of the second stage successively process and pass them to the voice recognition means 40. Here, the sound level output from the first sound signal output means 21 is detected by the first signal level detection means 121, and the sound level output from the second sound signal output means 22 is the second signal. It is detected by the level detection means 122 and passed to the learning timing determination means 130.

In the learning timing determining means 130, the first
Whether to learn the first signal level detecting means 121 or the second signal level detecting means 122 based on the result of the audio level detected by the signal level detecting means 121 and the second signal level detecting means 122. To decide.

Table 1 shows an example of setting the learning timing in the learning timing determining means 130.

[0084]

[Table 1]

In the example of Table 1, "learning timing determining means 1
Of the output signals of the first signal level detecting means 121 and the second signal level detecting means 122 input to 30,
When all the output signals except one (in this case, one output signal) are less than the threshold value A (= 50), and the remaining one output signal is the threshold value B (= 100) or more, The first-stage echo canceller means 111 or the second-stage echo canceller means 112, which receives the acoustic signal input to the first signal level detection means 121 or the second signal level detection means 122, or is shown in FIG. The learning of the first filter coefficient calculation means 141 or the second filter coefficient calculation means 142 is performed. "

According to the fourth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided. Further, by providing the learning timing determining means 130 for determining the learning timing of the echo canceller, the first acoustic signal output means 21 and the second acoustic signal output means 22 when recognizing the voice generated by the speaker. The acoustic signal output from the acoustic signal input means 30
It is possible to reduce the signal component that is mixed in, and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means, and three or more echo canceller means and signal level detection means are set. be able to.

FIG. 5 shows a block diagram of a speech recognition apparatus according to the fifth embodiment of the present invention.

As shown in FIG. 5, the speech recognition apparatus of the fifth embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10 are input, and the first acoustic signal output means 21 and the second acoustic signal output means are mixed from the output signal of the acoustic signal input means 30. It comprises an interfering sound removing section 100 for reducing the output signal component of 22 and a voice recognizing means 40 which receives the output signal of the interfering sound removing section 100 and recognizes the voice uttered by the speaker.

Further, in the disturbing sound removing section 100, one of the output signals of the acoustic signal generating means 10 and the output signal of the acoustic signal input means 30 are inputted, and the acoustic signal input means 3 is inputted.
First acoustic signal output means 2 mixed with 0 output signal
First filter coefficient calculation means 141 and second filter coefficient calculation means 142 for calculating the filter coefficient of a filter that acts to attenuate the output signal of one of the first and second acoustic signal output means 22; First filter coefficient calculation means 141 and second filter coefficient calculation means 1
First signal level detecting means 121 and second signal level detecting means 122 for detecting the level of the output signal of the acoustic signal generating means 10 input to 42, first signal level detecting means 121 and second signal Level detection means 122
Of the first filter coefficient calculation means 141 and the second filter coefficient calculation means 142 is determined from the signal level of the output signal, and the learning timing is determined. Learning timing determining means 130 and first filter coefficient calculating means 1
41 and the filter coefficient integrating means 15 for integrating the filter coefficients calculated by the second filter coefficient calculating means 142.
0, and a filtering unit 160 that performs a process of reducing the acoustic signal output component from the acoustic signal input unit 30 using the filter coefficient obtained by the filter coefficient integration unit 150.

The operation of the speech recognition apparatus of the fifth embodiment configured as above will be described. However,
Since the configuration other than the interference sound removing unit 100 is the same as that of the first embodiment, only the operation inside the interference sound removing unit 100 will be described.

The acoustic signal input from the acoustic signal 30 to the interfering sound removing section 100 is first passed to the first filter coefficient calculating means 141, and then output from the first acoustic signal output means 21 to the acoustic signal input means 30. The filter coefficient for reducing the mixed acoustic signal is calculated. The filter coefficient can be obtained, for example, by the principle of the adaptive filter described in the second embodiment.

In addition, the interfering sound removing unit 10 from the acoustic signal 30.
The acoustic signal input to 0 is also passed to the second filter coefficient calculation means 142, and a filter coefficient for reducing the acoustic signal output from the second acoustic signal output means 22 and mixed into the acoustic signal input means 30. Is calculated.

At this time, the sound level output from the first sound signal output means 21 is detected by the signal level detection means 121, and the sound level output from the second sound signal output means 22 is the signal level detection means 122. And is passed to the learning timing determining means 130. In the learning timing determining means 130, the first signal level detecting means 121
Then, the result of the voice level detected by the second signal level detecting means 122 is received, and it is determined whether or not to calculate the filter coefficient. Thus, the first filter coefficient calculation means 1
41 and the filter coefficient calculated by the second filter coefficient calculating means 142 are passed to the filter coefficient integrating means 150. The filter coefficient integration means 150 integrates two filter coefficients and converts them into one filter coefficient.

The first example of integrating filter coefficients is as follows.
Can be obtained by performing a convolution operation on the filter coefficients obtained by the filter coefficient calculation means 141 and the second filter coefficient calculation means 142.

The filter coefficient calculated by the filter coefficient integrating means 150 is passed to the filtering means 160. In the filtering unit 160, the output of the acoustic signal input unit 30 is filtered and output, so that the first
Acoustic signal output means 21 and second acoustic signal output means 22
The signal component that is output from and mixed into the acoustic signal input means 30 is reduced. According to the fifth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided and the filter is further added. By providing the means for calculating the coefficient and the means for integrating them, the first acoustic signal output means 21 and the second acoustic signal output means when recognizing the voice generated by the speaker are recognized. Two
It is possible to reduce the signal component in which the acoustic signal output from No. 2 is mixed in the acoustic signal input means 30 and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 6 shows a block diagram of a speech recognition apparatus according to the sixth embodiment of the present invention.

As shown in FIG. 6, the speech recognition apparatus according to the sixth embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And a noise suppression means 50 for performing noise suppression processing by inputting the output signal of the acoustic signal input means 30.
Of the first acoustic signal output means 21 and the second acoustic signal output means 22 which are input with the output signals of the noise suppression means 50 and the acoustic signal generation means 10 and are mixed into the output signals of the acoustic signal input means 30. It is composed of an interfering sound removing unit 100 that reduces an output signal component, and a voice recognizing unit 40 that receives an output signal of the interfering sound removing unit 100 and recognizes a voice uttered by a speaker.

The operation of the speech recognition apparatus of the sixth embodiment having the above configuration will be described. However,
Since the configuration other than the noise suppressing unit 50 is the same as that of the first embodiment, only the operation of the noise suppressing unit 50 will be described.

The noise suppression means 50 inputs the output signal of the acoustic signal input means 30, performs processing for removing the noise signal contained therein, and passes it to the interfering sound removing section 100. At this time, what is performed by the noise suppressing unit 50 is processing intended mainly to reduce environmental noise heard in a place where a speaker is present, and processing such as a spectral subtraction method is performed.

According to the sixth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 1
00 and the noise suppressor 50, the acoustic signals output from the first acoustic signal output means 21 and the second acoustic signal output means 22 are recognized when the voice generated by the speaker is recognized. It is possible to reduce the signal component mixed in the acoustic signal input means 30, further reduce the environmental noise existing in the place of the speaker, and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 7 shows a block diagram of a speech recognition apparatus according to the seventh embodiment of the present invention.

As shown in FIG. 7, the speech recognition apparatus of the seventh embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. , The first acoustic signal output means 21 and the second acoustic signal output means 22 which input the output signals of the acoustic signal input means 30 and the acoustic signal generation means 10 and are mixed in the output signals of the acoustic signal input means 30. Interference removal unit 100 for reducing the output signal component of
The noise suppression means 50 inputs the output signal of 00 to perform the noise suppression processing, and the voice recognition means 40 which inputs the output signal of the noise suppression means 50 and recognizes the voice uttered by the speaker.

The operation of the speech recognition apparatus of the seventh embodiment configured as above will be described. However,
Since the configuration other than the noise suppressing unit 50 is the same as that of the first embodiment, only the operation of the noise suppressing unit 50 will be described.

In the noise suppressing means 50, the interfering sound removing section 10
An output signal of 0 is input, a noise signal contained therein is removed, and the result is passed to the voice recognition means 40. At this time, what is performed by the noise suppressing unit 50 is processing intended mainly to reduce environmental noise heard in a place where a speaker is present, and processing such as a spectral subtraction method is performed.

According to the seventh embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided. Further, by providing the noise suppressing means 50, the first acoustic signal output means 21 and the second acoustic signal output means 22 are provided when the voice generated by the speaker is recognized.
It is possible to reduce the signal component in which the acoustic signal output from the device is mixed in the acoustic signal input means 30, further reduce the environmental noise existing at the place of the speaker, and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 8 shows a block diagram of a speech recognition apparatus according to the eighth embodiment of the present invention.

As shown in FIG. 8, the speech recognition apparatus of the eighth embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
A first acoustic signal output means 21 and a second acoustic signal output means 22 for outputting the output signal of the acoustic signal generation means 10 to the space as sound waves, and an acoustic signal input means 30 for inputting the voice uttered by the speaker. And the first noise suppressing means 51 for inputting the output signal of the acoustic signal input means 30 to perform noise suppressing processing, and the output signals of the first noise suppressing means 51 and the acoustic signal generating means 10 for inputting the acoustic signal. An interfering sound removing unit 100 that reduces output signal components of the first acoustic signal outputting unit 21 and the second acoustic signal outputting unit 22 mixed in the output signal of the input unit 30, and an output signal of the interfering sound removing unit 100. Second noise suppression means 52 for inputting and performing noise suppression processing
And a voice recognition means 40 for recognizing the voice uttered by the speaker by inputting the output signal of the second noise suppression means 52.

The operation of the speech recognition apparatus of the eighth embodiment constructed as above will be described. However,
Since the configuration other than the first noise suppressing means 51 and the second noise suppressing means 52 is the same as that of the first embodiment, only the operations of the first noise suppressing means 51 and the second noise suppressing means 52 will be described. To do.

First, in the first noise suppressing means 51, the output signal of the acoustic signal inputting means 30 is inputted, the noise signal contained therein is removed, and the interfering sound removing section 10 is carried out.
Pass to 0. The second noise suppressing means 52 still inputs the output signal of the interfering sound removing section 100, performs processing for removing the noise signal contained therein, and passes it to the voice recognizing means 40. At this time, the first noise suppressing means 51 and the second noise suppressing means 52 perform processing mainly for reducing the environmental noise heard in the place where the speaker is present, such as the spectral subtraction method. Process.
In addition, the first noise suppressing means 52 and the second noise suppressing means 5
The processes performed in 2 may be the same.

According to the eighth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 1
00, and by further providing the first noise suppressing means 51 and the second noise suppressing means 52, the first sound signal output means 21 and the second sound when the voice generated by the speaker is recognized. It is possible to reduce the signal component in which the acoustic signal output from the signal output unit 22 is mixed in the acoustic signal input unit 30, and further it is possible to reduce the environmental noise existing in the place of the speaker and improve the voice recognition performance. .

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 9 shows a block diagram of a speech recognition apparatus according to the ninth embodiment of the present invention.

As shown in FIG. 9, the speech recognition apparatus of the ninth embodiment has an acoustic signal generating means 10 for generating an acoustic signal.
, An acoustic signal output control means 60 for controlling the output of the acoustic signal generation means 10, and a first acoustic signal output means 21 and a second acoustic signal for outputting the output signal of the acoustic signal output control means 60 to the space as a sound wave. The output unit 22, the acoustic signal input unit 30 for inputting the voice uttered by the speaker, the output signal of the acoustic signal input unit 30, and the acoustic signal output control unit 60.
Of the first acoustic signal output means 21 and the second acoustic signal output means 21 which are mixed with the output signal of the acoustic signal input means 30.
The interfering sound removing unit 100 that reduces the output signal component of the acoustic signal outputting unit 22 and the voice recognizing unit 4 that receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker.
It is composed of 0 and 0.

The operation of the speech recognition apparatus of the ninth embodiment having the above configuration will be described. However,
Since the configuration other than the acoustic signal output control means 60 is the same as that of the first embodiment, only the operation of the acoustic signal output control means 60 will be described.

The acoustic signal output means 60 controls the output signal of the acoustic signal generation means 10. As an example of the control of the output signal, when the output signal of the acoustic signal generating means 10 is a monaural signal, it is converted into a stereo signal and is output to the first acoustic signal output means 21 and the second acoustic signal output means 22. May be output. In addition, the interfering sound removing unit 100
In order to improve the performance of the echo canceller or the like constituting the above, it is possible to output the signal of only one of the first acoustic signal output means 21 and the second acoustic signal output means 22.

According to the ninth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 1 is used.
No. 00 and the acoustic signal output control means 60 are provided, the performance of the interfering sound removing section can be improved, and the first acoustic signal output means 21 and the first acoustic signal output means 21 can be used when the voice generated by the speaker is recognized. Second acoustic signal output means 22
It is possible to reduce the signal component in which the acoustic signal output from the device is mixed in the acoustic signal input means 30, and it is possible to improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 10 shows a block diagram of a speech recognition apparatus according to the tenth embodiment of the present invention.

As shown in FIG. 10, the speech recognition apparatus according to the tenth embodiment has an acoustic signal generating means 10 for generating an acoustic signal and a transfer characteristic measuring means for generating a signal suitable for measuring the transfer characteristic. A signal generating means 70, and an acoustic signal output control means 60 for controlling so as to output one of the output signal of the acoustic signal generating means 10 and the output signal of the transfer characteristic measuring signal generating means 70 as necessary. , A plurality of first acoustic signal output means 21 and second acoustic signal output means 22 for outputting the output signal of the acoustic signal output control means 60 to the space as sound waves, and an acoustic signal input for inputting a voice uttered by a speaker. The first acoustic signal output means 2 in which the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal output control means 60 are input and mixed into the output signal of the acoustic signal input means 30. And a second acoustic signal output means 22
And an audio signal recognizing means 40 which receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker.

The operation of the speech recognition apparatus of the tenth embodiment constructed as above will be described. However, the configuration other than the transfer characteristic measuring signal generating means 70 is the ninth.
The acoustic signal output control means 6 is similar to that of the first embodiment.
Only 0 and the operation of the transfer characteristic measuring signal generating means 70 will be described.

First, the acoustic signal output control means 60 controls whether the acoustic signal generation means 10 outputs an acoustic signal or the transmission characteristic measurement signal generation means 70 outputs a transmission characteristic measurement signal.

When it is necessary to stabilize the operation in the interference sound removing section 100, a transfer characteristic measuring signal (for example, white noise or M) suitable for measuring the transfer characteristic from the transfer characteristic measuring signal generating means 70. Sequence signal) is output. The transfer characteristic measurement signal is the first acoustic signal output means 2
It is output via the first and second acoustic signal output means 22 and is input to the interfering sound removal unit 100 via the voice input means 30. The interfering sound removal unit 100 stabilizes the operation based on the input transfer characteristic measurement signal.

According to the tenth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided. Moreover, by providing the transfer characteristic measuring signal generating means for generating the signal for stabilizing the operation of the interference sound removing part 100, the signal is outputted from the first acoustic signal outputting means 21 and the second acoustic signal outputting means 22. The signal component mixed in the acoustic signal input means 30 can be reduced, and the voice recognition performance can be improved.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 11 shows a block diagram of a speech recognition apparatus according to the eleventh embodiment of the present invention.

As shown in FIG. 11, the speech recognition apparatus of the eleventh embodiment includes an acoustic signal generating means 10 for generating an acoustic signal, and an acoustic signal output control means 60 for controlling the output of the acoustic signal generating means 10. The first acoustic signal output means 21 and the second acoustic signal output means 22 that output the output signal of the acoustic signal output control means 60 to the space as sound waves, and the acoustic signal input means 30 that inputs the voice uttered by the speaker. Of the noise signal suppression means 50 for performing noise suppression processing by inputting the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal output control means 60, and the output signal of the noise suppression means 50 and the acoustic signal output control means 60. An interfering sound for inputting an output signal and reducing output signal components of the first acoustic signal output means 21 and the second acoustic signal output means 22 mixed in the output signal of the acoustic signal input means 30. And removed by section 100, disturbing sound removing unit 100
And a voice recognition means 40 for recognizing the voice uttered by the speaker.

The operation of the speech recognition apparatus of the eleventh embodiment constructed as above will be described. However, since the configuration other than the noise suppressing means 50 is the same as that of the ninth embodiment, only the operations of the acoustic signal output control means 60 and the noise suppressing means 50 will be described.

First, the acoustic signal output control means 60 controls the output signal of the acoustic signal generation means 10 and passes the control contents to the noise suppression means 50. The noise suppression unit 50 performs noise suppression processing from the output signal of the acoustic signal input unit 30, and outputs the result to the interference sound removal unit 100. At this time, the noise suppression unit 50 performs noise suppression based on the information of the signal output by the acoustic signal output control unit. For example, when no acoustic signal is output from both the first acoustic signal output means 21 and the second acoustic signal output means 22, it is possible to estimate noise and perform noise suppression processing.

According to the eleventh embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the noise canceller 100 and the noise canceller are used. By providing the suppressing means 50 and the acoustic signal output control means 60, when the voice generated by the speaker is recognized, the sound output from the first acoustic signal output means 21 and the second acoustic signal output means 22 is output. It is possible to reduce the signal component of the signal mixed in the acoustic signal input unit 30, further reduce the environmental noise existing at the location of the speaker, and improve the voice recognition performance.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 12 shows a block diagram of a speech recognition apparatus according to the twelfth embodiment of the present invention.

As shown in FIG. 12, the speech recognition apparatus of the twelfth embodiment includes an acoustic signal generating means 10 for generating an acoustic signal, and an acoustic signal output control means 60 for controlling the output of the acoustic signal generating means 10. The first acoustic signal output means 21 and the second acoustic signal output means 22 that output the output signal of the acoustic signal output control means 60 to the space as sound waves, and the acoustic signal input means 30 that inputs the voice uttered by the speaker. And the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal output control means 60 are input and mixed into the output signal of the acoustic signal input means 30. The noise canceling unit 100 that reduces the output signal component of the signal output unit 22, the noise canceling unit 100, the output signal of the noise canceling unit 100, and the output signal of the acoustic signal output control unit 60 are input. A suppression unit 50, the interference signal cancellation unit 10
The voice recognition means 40 receives the output signal of 0 and recognizes the voice uttered by the speaker.

The operation of the speech recognition apparatus of the twelfth embodiment constructed as above will be described. However, since the configuration other than the noise suppressing means 50 is the same as that of the ninth embodiment, only the operations of the acoustic signal output control means 60 and the noise suppressing means 50 will be described.

First, the acoustic signal output control means 60 controls the output signal of the acoustic signal generation means 10 and transfers the control content to the noise suppression means 50. The noise suppression unit 50 performs noise suppression processing on the output signal of the interference sound removal unit 100, and outputs the result to the voice recognition unit 40. At this time, in the noise suppressing means 50, the acoustic signal output controlling means 60
Noise suppression is performed based on the information of the signal output in. Thereby, for example, the first acoustic signal output means 21 and the second acoustic signal output means 21
It is possible to adjust the noise suppression amount according to the output level from the sound signal output means 22.

According to the twelfth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100, By providing the noise suppression means 50 and the acoustic signal output control means 60, the first acoustic signal output means 21 and the second acoustic signal output means 21 and the second acoustic signal output means 21 can be used when recognizing the voice generated by the speaker.
The acoustic signal output from the acoustic signal output means 22 can reduce the signal component mixed in the acoustic signal input means 30, and further reduce the environmental noise existing in the place of the speaker to improve the speech recognition performance. Can be improved.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 13 shows a block diagram of a speech recognition apparatus according to the thirteenth embodiment of the present invention.

As shown in FIG. 13, the speech recognition apparatus according to the thirteenth embodiment has a sound signal generating means 10 for generating a sound signal and a synthesized sound generating means 90 for generating a synthesized sound for realizing voice conversation. And a voice dialogue control means 80 which analyzes the voice recognition result and performs control for realizing voice dialogue,
The output signal of the acoustic signal generating means 10 and the synthetic sound generating means 90
Of the sound signal output control means 60 and the output signal of the voice interaction control means 80 are input, and the output signal of the sound signal output control means 60 is controlled by the sound wave. Acoustic signal output means 21 and second acoustic signal output means 22 for outputting to the space as
, The acoustic signal input means 30 for inputting the voice uttered by the speaker, the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10, and the acoustic signal input means 30.
First acoustic signal output means 21 mixed in the output signal of
From the interfering sound removing unit 100 that reduces the output signal component of the second acoustic signal output unit 22, and the voice recognizing unit 40 that receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker. It is configured.

The operation of the speech recognition apparatus of the thirteenth embodiment constructed as above will be described. However, since the configuration other than the synthetic sound generation unit 90, the voice interaction control unit 80, and the acoustic signal output control unit 60 is the same as that of the ninth embodiment, only these three operations will be described.

The acoustic signal output control means 60 controls whether the acoustic signal generation means 10 outputs an acoustic signal or the synthetic sound generated by the synthetic sound generation means 90. At this time, the voice interaction control means 80 outputs a response voice as a synthesized voice to the acoustic signal output means 60 based on the result of the voice recognition means 40, or outputs an acoustic signal from the acoustic signal generation means 10. Control whether to output. At this time, the acoustic signal output control means 60 instructs the synthetic sound generation means 90 to create a necessary synthetic sound, if necessary.

According to the thirteenth embodiment of the present invention as described above, in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22, the interfering sound removing section 100 is provided, Further, by providing the synthetic sound generation means 90 and the voice dialogue control means 80, the signal components output from the first acoustic signal output means 21 and the second acoustic signal output means 22 and mixed in the acoustic signal input means 30. Can be reduced, the voice recognition performance can be improved, and at the same time voice conversation can be realized.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 14 shows a block diagram of a speech recognition apparatus according to the 14th embodiment of the present invention.

As shown in FIG. 14, the speech recognition apparatus of the fourteenth embodiment has a sound signal generating means 10 for generating a sound signal and a synthesized sound generating means 90 for generating a synthesized sound for realizing a voice conversation. And a voice dialogue control means 80 which analyzes the voice recognition result and performs control for realizing voice dialogue,
The output signal of the acoustic signal generating means 10 and the synthetic sound generating means 90
Of the sound signal output control means 60 and the output signal of the voice interaction control means 80 are input, and the output signal of the sound signal output control means 60 is controlled by the sound wave. Acoustic signal output means 21 and second acoustic signal output means 22 for outputting to the space as
, The acoustic signal input means 30 for inputting the voice uttered by the speaker, the output signal of the acoustic signal input means 30 and the output signal of the acoustic signal generation means 10, and the acoustic signal input means 30.
First acoustic signal output means 21 mixed in the output signal of
And an interfering sound removing unit 100 that reduces the output signal component of the second acoustic signal output unit 22, and a voice recognizing unit 40 that receives the output signal of the interfering sound removing unit 100 and recognizes the voice uttered by the speaker. The display unit 95 is provided to provide information to the speaker for voice conversation.

The operation of the speech recognition apparatus of the fourteenth embodiment constructed as above will be described. However, since the configuration other than the display means 95 is the same as that of the thirteenth embodiment, only this operation will be described.

The display means 95 is means for providing information to the speaker in the form of video or characters by the voice dialogue control 80 for performing communication with the speaker by voice recognition and voice synthesis.
As an example of the specific operation content of the display unit 95, the voice recognition result performed by the voice recognition unit 40 is displayed to the speaker, or a map is displayed when the speaker requests the display of the map. Alternatively, the content of the acoustic signal generated by the acoustic signal generating means 10 may be provided to the speaker in text.

According to the fourteenth embodiment of the present invention as described above, the display means 9 is provided in the voice recognition device having the first acoustic signal output means 21 and the second acoustic signal output means 22.
5, the voice dialogue can be easily realized.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

FIG. 15 shows a block diagram of a speech recognition apparatus according to the 15th embodiment of the present invention.

As shown in FIG. 15, the speech recognition apparatus of the fifteenth embodiment has an acoustic signal generating means (audio) 11
An acoustic signal generating means (radio) 12 and a synthetic sound generating means 90 for generating a synthetic sound for realizing voice dialogue,
A voice interaction control unit 80 that analyzes the voice recognition result and performs control for realizing voice interaction, and a transfer characteristic measurement signal generation unit 7 that generates a signal suitable for measuring the transfer characteristic.
0, acoustic signal generation means 11 and acoustic signal generation means 1
2 output signal, the output signal of the synthetic sound generation means 90, the output signal of the transfer characteristic measurement signal generation means 70, and the output signal of the voice interaction control means 80 are input, and either signal is output as necessary. Signal output control means 6 for controlling as described above
0, the first acoustic signal output means 21, which outputs the output signal of the acoustic signal output control means 60 to the space as a sound wave, the second acoustic signal output means 22, and the third acoustic signal output means 23.
And the sound signal input means 30 for inputting the voice uttered by the speaker, the output signal of the sound signal input means 30, and the output signal of the sound signal output control means 60 are mixed and mixed into the output signal of the sound signal input means 30. The disturbing sound removing unit 10 that reduces the output signal components of the first acoustic signal output unit 21, the second acoustic signal output unit 22, and the third acoustic signal output unit 23 that are operating.
0, a voice recognition means 40 for recognizing a voice uttered by a speaker by inputting an output signal of the interfering sound removing section 100, and a display means 95 for displaying information necessary for voice conversation.

The operation of the speech recognition apparatus of the fifteenth embodiment configured as described above will be described.

In the acoustic signal output control means 60, the acoustic signal generating means 11, the acoustic signal generating means 12, the synthetic sound generating means 90, and the transfer characteristic measuring signal generating means 70 are used as necessary.
Is selected and output so as to be heard by the speaker through the first acoustic signal output means 21, the second acoustic signal output means 22, and the third acoustic signal output means 23.

The speaker inputs through the voice signal input means 30 and the output is passed to the noise suppressing means 50.
The noise suppressing unit 50 performs a process of suppressing noise generated in the environment where the speaker is present, and passes the output to the interfering sound removing unit 100. In the disturbance sound removing unit 100, the acoustic signals output from the first acoustic signal output means 21, the second acoustic signal output means 22, and the third acoustic signal output means 23 are acoustic signal input from the input acoustic signals. Processing for removing the signal component mixed in the means 30 is performed, and the result is passed to the voice recognition means 40. The voice recognition means 40 inputs the output signal of the interfering sound removing section 100, performs voice recognition processing, and passes the voice recognition result to the voice dialogue control means 80. The voice interaction control means 80 outputs the first acoustic signal output means 21, the second acoustic signal output means 22, and the third acoustic signal output means 23 to the acoustic signal output control means 60 based on the recognition result. The display unit 95 provides information to the speaker in accordance with the voice recognition result.

According to the fifteenth embodiment of the present invention as described above, the voice recognition having the first acoustic signal output means 21, the second acoustic signal output means 22 and the third acoustic signal output means 23. In the device, the first sound signal output unit 2 is provided by providing the disturbing sound removing unit 100, and further by providing the synthesized sound generating unit 90, the voice interaction control unit 80, and the display unit 95.
1, the second acoustic signal output means 22 and the third acoustic signal output means 23 can reduce the signal components mixed in the acoustic signal input means 30 to improve the voice recognition performance. At the same time, high-performance voice dialogue can be realized.

Similar to the first embodiment, the same processing can be performed when there are three or more acoustic signal output means.

Furthermore, in the first to fifteenth embodiments, the speech recognition apparatus according to the present invention has been described. However, the speech recognition method is used with each processing operation as a procedure, and the recognition performance is improved even when a plurality of acoustic signal output means are provided. Can be improved.

[0161]

As described above, the present invention has an effect that the voice recognition performance can be improved by adopting such a configuration that the signals are reduced even when a plurality of acoustic signals are output. .

[Brief description of drawings]

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

FIG. 2 is a block diagram of a voice recognition device according to a second embodiment of the present invention.

FIG. 3 is a block diagram of a voice recognition device according to a third embodiment of the present invention.

FIG. 4 is a block diagram of a voice recognition device according to a fourth embodiment of the present invention.

FIG. 5 is a block diagram of a voice recognition device according to a fifth embodiment of the present invention.

FIG. 6 is a block diagram of a voice recognition device according to a sixth embodiment of the present invention.

FIG. 7 is a block diagram of a voice recognition device according to a seventh embodiment of the present invention.

FIG. 8 is a block diagram of a voice recognition device according to an eighth embodiment of the present invention.

FIG. 9 is a block diagram of a voice recognition device according to a ninth embodiment of the present invention.

FIG. 10 is a block diagram of a voice recognition device according to a tenth embodiment of the present invention.

FIG. 11 is a block diagram of a voice recognition device according to an eleventh embodiment of the present invention.

FIG. 12 is a block diagram of a voice recognition device according to a twelfth embodiment of the present invention.

FIG. 13 is a block diagram of a voice recognition device according to a thirteenth embodiment of the present invention.

FIG. 14 is a block diagram of a voice recognition device according to a fourteenth embodiment of the present invention.

FIG. 15 is a block diagram of a speech recognition device according to a fifteenth embodiment of the present invention.

FIG. 16 is a block diagram of a block diagram of a typical echo canceller.

[Explanation of symbols]

10 Acoustic signal generating means 21 First acoustic signal output means 22 Second sound signal output means 23 Third acoustic signal output means 30 Sound signal input means 40 voice recognition means 50 Noise suppression means 51 First noise suppressing means 52 Second noise suppressing means 60 sound signal output control means 70 Transfer characteristic measuring signal generating means 80 Voice interaction control means 90 Synthetic sound generating means 95 display means 100 Interference remover 111 First Stage Echo Canceller Means 112 Second Stage Echo Canceller Means 121 First signal level detecting means 122 Second signal level detecting means 130 Learning timing determining means 141 First Filter Coefficient Calculation Means 142 Second filter coefficient calculation means 150 Filter coefficient integrating means 160 Filtering means 170 Adaptive filter means 200 External acoustic signal generation means

Claims (19)

[Claims] 1. An acoustic signal generation means for generating an acoustic signal, a plurality of acoustic signal output means for outputting the acoustic signal as a sound wave, an acoustic signal input means for inputting voice, and an output signal of the acoustic signal input means. And an interfering sound removing section for inputting an output signal of the acoustic signal generating means and reducing output signal components of the plurality of acoustic signal outputting means mixed in the output signal of the acoustic signal input means, and the interfering sound removing section. A voice recognition device comprising: a voice recognition unit that receives an output signal and recognizes the voice. 2. The acoustic signal generating means receives an output signal of a predetermined external acoustic signal generating means as an input, and outputs an acoustic signal to the plurality of acoustic signal outputting means. The voice recognition device described in. 3. The interfering sound removal section has a plurality of stages of echo canceller means for reducing the output signal component of the acoustic signal output means mixed in the output signal of the acoustic signal input means, and the echo canceller means of the frontmost stage is provided. The echo canceller means inputs the output signal of the acoustic signal input means, the echo canceller means of the other stage inputs the output signal of the echo canceller means of the preceding stage, and each echo canceller means has the acoustic signal generating means. 2. An output signal of one of the output signals of 1. is input, and the output signal component of the audio signal output means mixed in the output signal of the audio signal input means is reduced and output. The voice recognition device according to 2. 4. The interfering sound removing section detects a plurality of levels of the echo canceller means and a plurality of signal levels of the output signals of the acoustic signal generating means respectively inputted to the plurality of echo canceller means. The detection means and the output signals of the plurality of signal level detection means are input, and it is determined from the signal levels which one of the plurality of stages of echo canceller means is to perform the processing for improving the performance of the echo canceller. The speech recognition apparatus according to claim 3, further comprising a learning timing determining unit that determines a learning timing for improving the echo canceller performance. 5. The interfering sound removing section inputs any one of a plurality of output signals from the acoustic signal generating means and an output signal of the acoustic signal inputting means and mixes them with the output signal of the acoustic signal inputting means. A plurality of filter coefficient calculation means for calculating a filter coefficient of a filter that works to attenuate one output signal of the acoustic signal output means, and the acoustic signal input to each of the plurality of filter coefficient calculation means A plurality of signal level detecting means for respectively detecting the levels of the output signals from the generating means and the output signals of the plurality of signal level detecting means are input, and the filter coefficient of the filter coefficient calculating means is learned from the signal level. A learning timing determining means for determining which of the filter coefficient calculating means should be and determining a learning timing; The filter coefficient integrating means for integrating the filter coefficients calculated by the filter coefficient calculating means, the output signals of the filter coefficient integrating means and the acoustic signal input means are input, and the filter coefficients obtained by the filter coefficient integrating means are used. The speech recognition apparatus according to claim 1 or 2, further comprising a filtering unit that performs a process of reducing the acoustic signal output component. 6. The learning timing determining means is configured such that all output signals except one of the output signals of the plurality of signal level detecting means are less than a threshold value, and the remaining one output signal is less than the threshold value. When the threshold value is set separately, the echo canceller means or the filter coefficient calculation means, which receives the acoustic signal input to the signal level detection means as input, is learned. The voice recognition device according to Item 4 or 5. 7. The apparatus further comprises noise suppressing means for suppressing noise by inputting an output signal of the acoustic signal input means, wherein the interfering sound removing section outputs the output signal of the noise suppressing means and the acoustic signal generating means. 7. An output signal is input, and an output signal component of the acoustic signal output means mixed in the output signal of the acoustic signal input means is reduced.
The voice recognition device according to claim 1. 8. The apparatus further comprises noise suppression means for suppressing noise by inputting an output signal of the disturbing sound removing section, and the voice recognition means inputs an output signal of the noise suppression means and a speaker speaks. The recognized voice is recognized.
7. The voice recognition device according to any one of 1 to 6. 9. A first noise suppressing means for inputting an output signal of the acoustic signal input means to suppress noise, and a second noise suppressing means for inputting an output signal of the interfering sound removing section to suppress noise.
Noise suppressing means, and the interfering sound removing section inputs the output signal of the first noise suppressing means and the output signal of the acoustic signal generating means to the output signal of the first noise suppressing means. The mixed output signal component of the acoustic signal output means is reduced, and the voice recognition means inputs the output signal of the second noise suppression means to recognize the voice generated by the speaker. The voice recognition device according to any one of claims 1 to 6. 10. An acoustic signal output control means for controlling an output of the acoustic signal generation means, wherein the acoustic signal output control means inputs an output signal of the acoustic signal generation means, and outputs the plurality of acoustic signals. The signal output to the output means and the interfering sound removing section is controlled.
10. The voice recognition device according to any one of 1 to 9. 11. A transfer characteristic measuring signal generating means for generating a signal suitable for measuring a transfer characteristic, wherein the acoustic signal output control means and the output signal of the acoustic signal generating means and the 11. The voice recognition device according to claim 10, wherein the voice recognition device is controlled so as to output one of the output signals of the transfer characteristic measuring signal generating means. 12. The speech recognition apparatus according to claim 11, wherein the signal output from the transfer characteristic measuring signal generating means is either white noise or an M-sequence signal. 13. The apparatus further comprises noise suppressing means for suppressing noise from information on the output signal of the acoustic signal input means and the signal output from the acoustic signal output control means, and the interfering sound removing section includes the noise reducing means. 11. The output signal of the suppressing means is input and processed.
The voice recognition device according to any one of 2 above. 14. A noise suppressing unit for suppressing noise from information about an output signal of the interfering sound removing unit and a signal output from the acoustic signal output control unit, and the voice recognizing unit includes the noise suppressing unit. 14. The voice recognition device according to claim 10, wherein the voice signal uttered by the speaker is recognized by inputting the output signal of the means. 15. The system further comprises a synthetic sound generation unit that creates a synthetic sound to be heard by a speaker, and a voice dialogue control unit that analyzes the voice recognition result and performs control for realizing a voice dialogue. The signal output control means inputs the output signal of the acoustic signal generation means, the output signal of the synthesized sound generation means, and the output signal of the voice interaction control means, and outputs the output signal of the acoustic signal generation means and the 2. Outputting one of the output signals of the synthetic sound generation means.
The voice recognition device according to any one of 0 to 14. 16. The voice recognition device according to claim 15, further comprising display means for displaying the contents determined by the voice interaction control means. 17. The voice recognition device according to claim 1, wherein the signal output from the acoustic signal generating means is a stereo signal. 18. The voice recognition device according to claim 1, wherein a plurality of signals output from the acoustic signal generating means have the same content. 19. An acoustic signal generation procedure for generating an acoustic signal, a plurality of acoustic signal output procedures for outputting the acoustic signals as sound waves, an acoustic signal input procedure for inputting voice, and an output in the acoustic signal input procedure. A signal and an output signal in the acoustic signal generation procedure, and an interference sound removal procedure for reducing output signal components in the plurality of acoustic signal output procedures mixed in the output signal in the acoustic signal input procedure; A voice recognition method comprising: inputting an output signal of a sound removal procedure and performing a voice recognition procedure for recognizing the voice.