JP2004309536A - Speech processing unit - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain a speech processing unit reliably estimating whether a speaker speaks according to a speech signal inputted from a microphone. <P>SOLUTION: The speech processing unit 10 when inputting a 1st speech signal and a 2nd speech signal from a 1st microphone 12A and a 2nd microphone 12B arranged nearly symmetrically to a specified speaker calculates voice power as the power of the sum signal of the 1st and 2nd speech signals and noise power as the power of a difference signal generated by subtracting the 2nd speech signal from the 1st speech signal, and a speaking/non-speaking state estimation part 32 estimates whether the speaker speaks according to whether the voice power is larger or smaller than the noise power which is not influenced by speaker's speaking. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a voice processing device applied to, for example, a voice recognition device or a communication device.
[0002]
[Prior art]
For example, some navigation devices and audio devices mounted on vehicles such as automobiles are provided with a voice recognition device that inputs and recognizes the voice of an occupant from a microphone and performs various processes based on the recognition content. Have been. Some vehicles have a hands-free communication device for the driver to talk without having to hold a microphone or the like (for example, the main body of a mobile phone) while driving.
[0003]
Some of such speech recognition devices and communication devices include two microphones and realize a noise canceling function for removing voices other than voices emitted by an occupant (speaker) (for example, see Patent Document 1). reference). Specifically, an audio signal (output signal) corresponding to a speaker's voice is extracted by subtracting a noise signal input from the other microphone and processed by the noise cancellation filter from an audio signal input from one microphone. And output it. An adaptive filter is used as the noise canceling filter, and the filter coefficient is updated so that the power of the output signal as the difference signal is minimized.
[0004]
However, in the configuration in which the noise cancellation filter constantly updates the filter coefficient, when the voice of the speaker is loud relative to the noise, a part of the signal corresponding to the voice of the speaker is also removed (canceled), and the output signal is large. Distortion occurs. For this reason, in the above-described speech recognition device and the communication device, when the speaker speaks, the speaker presses the speech switch to fix the filter coefficient of the noise cancellation filter to correspond to the speaker's voice. The output signal obtained in this manner is prevented from being removed from a part of the output signal. Thereby, distortion of the output signal is prevented.
[0005]
Further, there has been known a voice recognition device or a communication device as described above that realizes a noise canceling function by a configuration different from the above configuration (for example, see Patent Document 2). The vehicle audio input device described in Patent Document 2 is provided with a plurality of microphones corresponding to each seat of the vehicle, and based on a moving average value of an audio signal level input from each microphone in a past predetermined time. A threshold value (power corresponding to noise) is set, and when an input signal level in any one of the microphones exceeds the corresponding threshold value, it is determined that a voice by any occupant has been input, and a specific occupant (seat) is determined. When the input signal level from the microphone corresponding to the other occupant is higher than the input signal level from the microphone corresponding to another occupant, it is determined that the specific occupant has uttered and an output signal is output. I have.
[0006]
The output signal is a signal obtained by subtracting an input signal from a microphone corresponding to another occupant from an input signal from a microphone corresponding to a specific occupant (seat) and removing noise. That is, this configuration does not include a noise canceling filter that is an adaptive filter, and basically receives input from a microphone corresponding to another occupant (a microphone that is sufficiently spaced from a microphone corresponding to a specific occupant). Is a noise signal. With this configuration, in this configuration, without operating the utterance switch, it is determined whether or not the specific occupant has uttered, and when it is determined that the specific occupant has uttered, the microphone corresponding to the specific occupant is used. An output signal obtained by removing noise from the input signal can be output to the speech recognition device.
[0007]
[Patent Document 1]
JP 2000-148200 A
[Patent Document 2]
JP-A-11-65586
[0008]
[Problems to be solved by the invention]
However, the former configuration has a problem in that the speaker must operate the utterance switch every time the utterance is made, and the operation is complicated.
[0009]
On the other hand, in the latter configuration, since a plurality of microphones are arranged corresponding to different occupants, in other words, since there is only one microphone corresponding to a specific occupant and noise cannot be separated, the criterion for determining the presence or absence of speech is determined. There is a problem that the threshold value is likely to fluctuate depending on the size of the voice and the size of the noise, and the accuracy of determining the presence or absence of speech is poor. Therefore, for example, when noise occurs near the microphone corresponding to a specific occupant (noise giving a large input to the microphone), the noise cannot be removed by a small input signal to another microphone, and There was a high possibility that a wrong judgment was made. In particular, since the input signal level compared with the threshold (power based on the moving average value) at the time of determining the presence or absence of speech is an instantaneous value, this problem becomes remarkable.
[0010]
SUMMARY OF THE INVENTION It is an object of the present invention to provide an audio processing device capable of reliably estimating the presence or absence of an utterance by a speaker based on an audio signal input from a microphone in consideration of the above fact.
[0011]
[Means for Solving the Problems]
In order to achieve the above object, an audio processing apparatus according to the present invention is characterized in that first audio signals respectively input from a first microphone and a second microphone arranged substantially symmetrically with respect to a specific speaker. An audio processing device for processing a second audio signal, the power of an addition signal obtained by adding the first audio signal and the second audio signal, and a difference signal obtained by subtracting a second audio signal from the first audio signal. Utterance state estimating means for estimating the presence / absence of utterance by the speaker by comparing the power with the utterance.
[0012]
In the audio processing device according to the first aspect, the first audio signal is input from the first microphone, and the second audio signal is input from the second microphone. Since the first and second microphones are arranged substantially symmetrically with respect to the speaker, the first and second audio signals are signals having substantially the same phase and loudness as to the voice emitted by the speaker. The voice other than the voice uttered by the speaker, that is, noise differs depending on the relative position to the source.
[0013]
When the first and second audio signals are input, the utterance state estimating means calculates the sum of the power of the first audio signal and the second audio signal (hereinafter referred to as audio power) and the first audio signal. By comparing the power of the difference signal obtained by subtracting the second audio signal (hereinafter referred to as noise power), the presence or absence of the utterance of the speaker is estimated (determined, determined). Needless to say, the audio power and the noise power may not be directly compared, but may be a power obtained by appropriately processing one or both of them, for example.
[0014]
Specifically, the voice power is the power of the added signal of the first and second voice signals, both of which are only noise when the speaker is not uttering, so the first and second voices, which are both noise only, The ratio of the difference signal power of the audio signal to the noise power, which is the power of the difference signal, is small. On the other hand, when the speaker utters a voice, the voice power, which is the power of the addition signal, includes the power of the signal corresponding to the voice of the speaker, and is therefore the power of the difference signal, which corresponds to the voice of the speaker. The ratio of the signal to the noise power not including the power is sufficiently large. Since the noise power has little difference in phase and magnitude depending on the presence or absence of the speaker's utterance, for example, by comparing a threshold obtained by multiplying this noise power by an appropriate coefficient with the voice power, the utterance by the speaker is determined. Is estimated.
[0015]
Then, based on the first and second audio signals from the first and second microphones respectively arranged corresponding to the specific speakers, that is, based on the signal including the information of the spatial sound, the presence or absence of the utterance is determined. Since the noise power that is hardly affected is separated and the presence or absence of the utterance is estimated based on the noise power, the presence or absence of the utterance can be reliably estimated regardless of the position of the noise source. Moreover, since the presence / absence of speech is estimated by comparing the voice power and the noise power, the possibility of erroneous estimation is significantly reduced as compared with the case where the presence / absence of speech is estimated using instantaneous values.
[0016]
As a result, control based on the presence or absence of speech can be performed without operating the speech switch as in the related art. Therefore, for example, an audio signal is output to a voice input device (speech recognition device, communication device, or the like) only when it is estimated that there is utterance, or the filter coefficient of the noise cancellation filter is changed when it is estimated that there is utterance. It is possible to perform control such as dropping and fixing.
[0017]
As described above, in the voice processing device according to the first aspect, it is possible to reliably estimate the presence or absence of the utterance by the speaker based on the voice signal input from the microphone.
[0018]
According to another aspect of the present invention, there is provided a speech processing apparatus comprising: a first microphone and a second microphone arranged substantially symmetrically with respect to a specific speaker; An audio processing device for processing an audio signal and a second audio signal, wherein the audio power calculation means obtains audio power by calculating the power of an addition signal obtained by adding the first audio signal and the second audio signal; Noise power calculating means for calculating the power of a difference signal obtained by subtracting the second audio signal from the first audio signal to obtain noise power; comparing the difference between the audio power and the noise power with the noise power; Utterance state estimating means for estimating the presence / absence of the utterance.
[0019]
In the audio processing device according to the second aspect, the first audio signal is input from the first microphone, and the second audio signal is input from the second microphone. Since the first and second microphones are arranged substantially symmetrically with respect to the speaker, the first and second audio signals are signals having substantially the same phase and loudness as to the voice emitted by the speaker. The voice other than the voice uttered by the speaker, that is, noise differs depending on the relative position to the source.
[0020]
When the first and second audio signals are input, the audio power calculation means calculates the power of the added signal obtained by adding the first audio signal and the second audio signal to obtain the audio power, and the noise power calculation means obtains the audio power. The noise power is obtained by calculating the power of the difference signal obtained by subtracting the second audio signal from the first audio signal. Then, the utterance state estimating unit estimates (determines and determines) whether or not the utterer has uttered by comparing the difference between the voice power and the noise power (hereinafter, referred to as pseudo utterance power) with the noise power.
[0021]
Specifically, the pseudo utterance power, which is the difference between the audio power and the noise power, substantially corresponds to the power of the entire audio signal minus the power of the noise signal, and when the speaker does not speak, the utterance power is reduced. Since the power of the signal corresponding to the speaker's voice is not included, the ratio to the noise power is small, and when the speaker speaks, the ratio to the noise power is sufficient because the power of the signal corresponding to the speaker's voice is included. large. Since the noise power has almost no difference in phase and magnitude depending on the presence or absence of the utterance of the speaker, for example, by comparing a threshold obtained by multiplying the noise power by an appropriate coefficient with the pseudo utterance power, The presence or absence of speech is estimated.
[0022]
Then, based on the first and second audio signals from the first and second microphones respectively arranged corresponding to the specific speakers, that is, based on the signal including the information of the spatial sound, the presence or absence of the utterance is determined. Since the noise power that is hardly affected is separated and the presence or absence of the utterance is estimated based on the noise power, the presence or absence of the utterance can be reliably estimated regardless of the position of the noise source. Moreover, since the presence or absence of the utterance is estimated by comparing the pseudo utterance power and the noise power, the possibility of erroneous estimation is significantly reduced as compared with the case where the presence or absence of the utterance is estimated using the instantaneous value.
[0023]
As a result, control based on the presence or absence of speech can be performed without operating the speech switch as in the related art. Therefore, for example, an audio signal is output to a voice input device (speech recognition device, communication device, or the like) only when it is estimated that there is utterance, or the filter coefficient of the noise cancellation filter is changed when it is estimated that there is utterance. And can be fixed.
[0024]
As described above, in the voice processing device according to the second aspect, it is possible to reliably estimate the presence or absence of the utterance by the speaker based on the voice signal input from the microphone.
[0025]
It should be noted that the power in claims 1 and 2 is not limited to the mean square value of the signal in a predetermined time (predetermined movement time). included.
[0026]
BEST MODE FOR CARRYING OUT THE INVENTION
An audio processing device 10 according to an embodiment of the present invention will be described with reference to FIG. FIG. 1 is a schematic block diagram showing the overall configuration of the audio processing device 10. As shown in the figure, the audio processing device 10 is electrically connected to the microphone array 12 and configured to process audio signals input from the microphone array 12.
[0027]
The microphone array 12 includes a first microphone 12A and a second microphone 12B. The first microphone 12A and the second microphone 12B are arranged substantially symmetrically with respect to a specific speaker in front of the specific speaker. ing. That is, the voice uttered by a specific speaker is input to the first microphone 12A and the second microphone 12B with substantially the same phase and magnitude.
[0028]
The first microphone 12A and the second microphone 12B are electrically connected to the audio processing device 10 via A / D converters 14 and 16, respectively, and the audio processing device 10 has an A / D converter. In this configuration, audio signals digitized by 14 and 16 are input. Hereinafter, the term “connection” means an electrical connection.
[0029]
The audio processing device 10 includes an adding unit 18 and a subtracting unit 20. The adder 18 is connected to the A / D converters 14 and 16, respectively, and adds the first audio signal input from the A / D converter 14 and the second audio signal input from the A / D converter 16. And output. On the other hand, the subtraction unit 20 is configured to calculate and output the difference (absolute value) between the first audio signal input from the A / D converter 14 and the second audio signal input from the A / D converter 16. .
[0030]
Further, the adding unit 18 is connected to a band-pass filter (hereinafter, referred to as BPF) 22, and the subtracting unit 20 is connected to a BPF 24. The BPFs 22 and 24 are configured to cut out signal components other than the frequency band of the human voice from the signals input from the adding unit 18 and the subtracting unit 20, respectively, and output a voice signal in the frequency band of the human voice.
[0031]
Further, the BPF 22 is connected to an audio power calculator 26 as audio power calculator, and the BPF 24 is connected to a noise power calculator 28 as noise power calculator. The audio power calculator 26 calculates the power of the output signal of the BPF 22 for a predetermined time and outputs the calculated power as audio power. The noise power calculator 28 calculates the power of the output signal of the BPF 24 for the predetermined time. Output as noise power. These powers can be, for example, an average value, a root-mean-square value, or a power spectrum value in the above-mentioned predetermined time. In the present embodiment, the power is a root-mean-square value, and is constantly updated. .
[0032]
Furthermore, the audio power calculation unit 26 and the noise power calculation unit 28 are connected to the subtraction unit 30, respectively. The subtraction unit 30 is configured to calculate a difference between the voice power input from the voice power calculation unit 26 and the noise power input from the noise power calculation unit 28 and output the difference as pseudo speech power. The subtraction unit 30 is connected to an utterance presence / absence estimation unit 32 as utterance state estimation means. The utterance presence / absence estimation unit 32 is also connected to a noise power calculation unit 28.
[0033]
Then, the utterance presence / absence estimation unit 32 compares the pseudo utterance power input from the subtraction unit 30 with the noise power input from the noise power calculation unit 28, and estimates (determines / determines) the presence / absence of utterance by the speaker. It has been. In the present embodiment, the utterance presence / absence estimating unit 32 estimates that the utterance has been made by the utterer when the magnitude of the pseudo utterance power exceeds three times the magnitude of the noise power, and outputs the utterance presence signal T. It is supposed to.
[0034]
The utterance presence / absence estimation unit 32 is connected to a voice input device 34 and a noise canceling filter 36 constituting a voice output device as necessary. The voice input device 34 is, for example, a communication device such as a voice recognition device or a mobile telephone device that forms a navigation device or the like. The noise canceling filter 36 is an adaptive filter that removes noise from the audio signal output to the audio input device 34, and is capable of changing a filter coefficient. The utterance presence / absence estimating unit 32 is configured to output the utterance presence signal T to the voice input device 34 as trigger information and to the noise cancellation filter 36 as a control signal.
[0035]
In the BPFs 22 and 24, a gain difference between the addition signal input to the audio power calculation unit 26 and the difference signal input to the noise power calculation unit 28 easily occurs within the range of the frequency band described above. It is designed to pass signals in bands (each band may be different), and is configured to easily detect pseudo speech power (voice power) and noise power.
[0036]
Next, the operation of the present embodiment will be described.
[0037]
As shown in FIG. 1, when the voice S of the speaker reaches the microphone array 12 from the front and the noise N arrives from a position other than the front, the voice S and the noise N1 are input to the first microphone 12A. The voice S and the noise N2 are input to the second microphone 12B.
[0038]
Then, the first audio signal [S + N1] input from the first microphone 12A is digitally converted by the A / D converter 14, and then output to the adding unit 18 and the subtracting unit 20 of the audio processing device 10, respectively. The second audio signal (S + N2) input from the second microphone 12B is digitally converted by the A / D converter 16 and then output to the adding unit 18 and the subtracting unit 20, respectively.
[0039]
The adder 18 outputs an addition signal [S + N '] (= 2S + N1 + N2) obtained by adding the first audio signal and the second audio signal. As described above, the addition signal includes the signal component [S] corresponding to the voice of the speaker and the noise component [N ′] corresponding to the noise. On the other hand, the subtractor 20 outputs a difference signal [N ″] (= N1−N2) obtained by calculating a difference between the first audio signal and the second audio signal. As described above, the difference signal does not include the signal component [S] corresponding to the speaker's voice, but includes only the noise component [N ″] similar to the noise component [N ′].
[0040]
In the addition signal, the band components other than the predetermined frequency band are cut by the BPF 22 and input to the audio power calculation unit 26. The audio power calculator 26 outputs the audio power [P (S + N ′)] obtained by calculating the mean square of the added signal for a predetermined time to the subtractor 30.
[0041]
On the other hand, the BPF 24 cuts out the band components other than the predetermined frequency band from the subtraction signal, and inputs the signal to the noise power calculator 28. The noise power calculation unit 28 outputs the noise power [P (N ″)] obtained by calculating the mean square of the added signal for a predetermined time to the subtraction unit 30 and the utterance presence / absence estimation unit 32.
[0042]
In the subtraction unit 30 to which the voice power and the noise power are input, the pseudo utterance power [P (S ′)] (= P (S + N ′) − P (N ″)) obtained by subtracting the noise power from the voice power. Is output to the utterance presence / absence estimation unit 32. Then, the utterance presence / absence estimation unit 32 compares the pseudo utterance power [P (S ′)] with the noise power [P (N ″)], and finds that P (S ′)> 3 × P (N ″). When it is established, it is estimated that the utterance has been made by the speaker, and the utterance presence signal T is output.
[0043]
On the other hand, when there is no utterance, since the voice S is not input from the microphone array 12 to the voice processing device 10, the output of the adder 18, that is, the input to the voice power calculator 26 is the addition signal [N '] (= N1 + N2). ), And the output of the audio power calculator 26 is the audio power [P (N ′)]. The output of the subtraction unit 20, that is, the input to the noise power calculation unit 28 is the same difference signal [N ″] as in the case where the speech is present, and the output of the noise power calculation unit 28 is the noise power [P (N '')].
[0044]
As a result, the output of the subtraction unit 30 is the pseudo speech power [P (S ″)] (= P (N ′) − P (N ″)), and this pseudo speech power is three times the noise power. It is estimated that there is no utterance without exceeding.
[0045]
As described above, the voice processing device 10 determines the magnitude of the pseudo-utterance power that can include the component corresponding to the voice S generated by the utterance based on the noise power that is not affected by the presence or absence of the utterance. Presence / absence can be reliably estimated. Then, based on the first and second audio signals from the first microphone 12A and the second microphone 12B respectively arranged corresponding to the specific speaker, that is, the signal of the utterance based on the signal including the information of the spatial sound. Since the presence / absence is estimated, the presence / absence of the utterance can be more reliably estimated regardless of the position of the noise source. Furthermore, since the noise power is compared with the pseudo-utterance power to estimate the presence or absence of speech, the possibility of erroneous estimation is significantly reduced as compared with the case where the presence or absence of speech is estimated using instantaneous values. .
[0046]
Then, since the utterance presence signal T, which is the estimation result, is output to the voice input device 34 as trigger information, the voice input device 34 determines whether or not the utterance is made by the utterer without operating the utterance switch. It becomes possible to detect. Similarly, in the audio output device having the noise canceling filter 36, it is possible for the speaker to detect the presence or absence of the utterance without operating the utterance switch. Can be controlled such as changing the filter coefficient.
[0047]
As described above, in the voice processing device 10 according to the present embodiment, it is possible to reliably estimate the presence or absence of the utterance by the speaker based on the voice signal input from the microphone array 12.
[0048]
Next, an example in which the voice processing device 10 is applied to a voice recognition navigation system 50 that is mounted on a vehicle such as an automobile and is an embodiment of the voice input device 34 will be described with reference to FIG.
[0049]
As shown in FIG. 2, the voice recognition navigation system 50 is a navigation device that is controlled as a control command using a voice emitted by a vehicle occupant (in the present embodiment, a speaker who is a driver), and is connected to each other. It includes a voice processing unit 52, a voice output unit 54, and a voice recognition navigation device 56.
[0050]
The audio processing unit 52 has the same configuration as the audio processing device 10 except that the BPFs 22 and 24 are connected to the addition unit 18 and the subtraction unit 20 of the audio output unit 54, respectively. That is, the addition unit 18 and the subtraction unit 20 are shared by the audio processing unit 52 and the audio output unit 54.
[0051]
The audio output unit 54 includes a microphone array 12 including a first microphone 12A and a second microphone 12B, A / D converters 14 and 16, an adding unit 18, and a subtracting unit 20, which are The connection is made in exactly the same way as in the above embodiment. The first microphone 12A and the second microphone 12B are arranged at predetermined intervals (for example, 100 mm) symmetrically with respect to the driver near an instrument panel in front of a driver's seat and a front end of a roof in a vehicle cabin.
[0052]
The adder 18 is connected to the BPF 22 of the audio processor 52 as described above, and is also connected to the subtractor 58 via the delay processor 57. On the other hand, the subtraction unit 20 is connected to the BPF 24 of the audio processing unit 52 and the signal input unit of the noise cancellation filter 36 as described above.
[0053]
The noise canceling filter 36 is an adaptive filter as described above, and its output is connected to the subtractor 58. The subtractor 58 includes an adder signal of the adder 18 delayed by the delay processor 57 in accordance with the signal processing time of the noise cancel filter 36, and a noise signal input from the noise cancel filter 36 (a signal obtained by processing the difference signal). ) Is calculated, and a pseudo speech signal which is a voice signal obtained by the calculation is output.
[0054]
The output of the subtractor 58 is connected to the speech recognition navigation device 56 and to the control input of the noise cancellation filter 36. That is, the noise cancellation filter 36 is configured to feed back an audio signal (pseudo utterance signal). Normally, the noise cancellation filter 36 outputs the noise signal with the minimum power of the audio signal so that the filter coefficient is reduced. Is updated.
[0055]
The control input unit of the noise cancellation filter 36 is connected to the utterance presence / absence estimation unit 32 of the audio processing unit 52, and the utterance presence signal T is input as a control signal. The noise canceling filter 36 is configured to stop updating the filter coefficient while the speech presence signal T is being input, and maintain the filter coefficient immediately before the speech presence signal T is input.
[0056]
Further, the speech presence / absence estimation unit 32 of the speech processing unit 52 is connected to the speech recognition navigation device 56, and outputs the speech presence signal T as trigger information. When the speech recognition signal T is input, the voice recognition navigation device 56 receives a voice signal input from the subtractor 58 of the voice output unit 54 as a voice signal of the speaker, and performs a voice recognition process based on the voice signal. And the recognition result is used as a control command. That is, scrolling and zooming of the map screen, various searches, and the like (switching of the hierarchical virtual control panel) are performed based on the voice uttered by the speaker.
[0057]
In the speech recognition navigation system 50 described above, when the speaker's voice S reaches the microphone array 12 from the front and the noise N from other than the front, the voice S and the noise N1 are transmitted to the first microphone 12A. , And the voice S and the noise N2 are input to the second microphone 12B.
[0058]
Then, as in the case of the audio processing device 10, the adding unit 18 outputs the addition signal [S + N '], and the subtracting unit 20 outputs the difference signal [N "]. The noise cancellation filter 36 to which the difference signal [N ″] has been input processes this and outputs a noise signal [N ′ ″]. On the other hand, the delay processing unit 57 to which the addition signal [S + N ′] is input delays the addition signal according to the processing time of the difference signal [N ″] by the noise cancellation filter 36 and outputs the result.
[0059]
Then, in the audio output unit 54, the subtracter 58 subtracts the noise signal [N ′ ″] input from the noise cancellation filter 36 from the addition signal [S + N ′] input from the delay processing unit 57 to reduce the noise component. A pseudo speech signal [S '], which is a removed speech signal, is output. At this time, the noise cancellation filter 36 updates the filter coefficient so as to minimize the pseudo speech signal, in other words, to remove as much as possible a noise component that changes with time.
[0060]
On the other hand, the speech processing unit 52 to which the addition signal and the subtraction signal are input compares the pseudo speech power [P (S ′)] with the noise power [P (N ″)] in the speech presence / absence estimation unit 32. When P (S ′)> 3 × P (N ″) holds, the utterance presence / absence estimating unit 32 estimates that the utterer has made an utterance, and outputs an utterance presence signal T.
[0061]
When the utterance presence signal T is input, the noise cancellation filter 36 of the audio output unit 54 fixes the filter coefficient to the filter coefficient immediately before the input. Thus, after the speech processing unit 52 estimates that the utterance is present, the output from the subtracter 58 maintains the function of removing as much as possible the noise component corresponding to the noise generated immediately before the output of the utterance signal T. On the other hand, the problem that a part of the signal component [S] corresponding to the speaker's voice is removed by the operation of the noise canceling filter 36 to minimize the pseudo utterance signal [S '] is solved.
[0062]
That is, the subtractor 58 outputs a good pseudo-utterance signal [S ′] having a signal component [S] corresponding to the speaker's voice as a main component and having little distortion. The pseudo speech signal is input to the speech recognition navigation device 56. At this time, the speech recognition navigation device 56 receives the pseudo speech signal as the speech signal of the speaker since the speech presence signal T is input. Further, the speech recognition navigation device 56 performs a speech recognition process based on the utterance signal, and uses the recognition result as a control command.
[0063]
As described above, in the voice recognition navigation system 50, the voice processing unit 52 (the utterance presence / absence estimation unit 32) reliably estimates the presence or absence of the utterance by the speaker based on the input information from the microphone array 12. Does not need to operate the utterance switch before speaking. For this reason, when there is an utterance (estimated) by the speaker, control for fixing the filter coefficient of the noise cancel filter 36 can be performed without operating the utterance switch, and the signal corresponding to the utterer's voice can be controlled. Partial removal prevents a large distortion from occurring in the pseudo speech signal. That is, in the voice recognition navigation system 50, a good pseudo utterance signal with little distortion can be input to the voice recognition navigation device 56.
[0064]
Further, since the speech presence signal T is also input as trigger information to the speech recognition navigation device 56, the speech recognition navigation device 56 performs speech recognition based on a pseudo speech signal input when there is no speech by the speaker. And there is no risk of malfunction. This function is realized without operating the speech switch.
[0065]
In the above-described embodiment, only the main components have been described for the sake of simplicity and easy understanding. However, each of the adding units, the adder or the subtracting unit, and the attenuating unit or the Needless to say, it is possible to improve the accuracy of extracting a pseudo speech signal (the accuracy of removing noise and the like) by providing an amplifier. Further, in the voice recognition navigation system 50 mounted on the vehicle, for example, an adaptive filter or the like for removing the voice of the in-vehicle audio device may be added to the above configuration.
[0066]
Further, in the above embodiment, the speech processing device 10 and the speech processing unit 52 are configured to compare the pseudo speech power and the noise power to estimate the presence or absence of speech by the speaker, but the present invention is not limited to this. Instead, for example, a configuration may be adopted in which the presence or absence of the utterance by the speaker is estimated by comparing the voice power and the noise power.
[0067]
Further, in the above-described embodiment, an example in which the voice processing device 10 is applied to the voice recognition navigation system 50 has been described. However, the present invention is not limited to this. The present invention can be applied to various communication devices such as various devices, a hands-free communication system mounted on a vehicle, a transceiver device, a so-called video conference system, and a simple telephone device. Further, the voice processing device 10 may be applied to a navigation device, an audio device, an air conditioner, a hands-free communication system, a voice recognition device for integrated control of in-vehicle devices, and the like mounted on a vehicle or the like.
[0068]
【The invention's effect】
As described above, the speech processing device according to the present invention has an excellent effect that the presence or absence of speech by a speaker can be reliably estimated based on a speech signal input from a microphone.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a schematic configuration of an audio processing device according to an embodiment of the present invention.
FIG. 2 is a block diagram showing an example in which the speech processing device according to the embodiment of the present invention is applied to a speech recognition navigation system.
[Explanation of symbols]
10 Audio processing device
12A first microphone (first microphone)
12B 2nd microphone (2nd microphone)
26 Audio power calculation unit (Audio power calculation means)
28 Noise Power Calculator (Noise Power Calculator)
32 Speech presence / absence estimation unit (speech state estimation means)

Claims (2)

An audio processing device for processing a first audio signal and a second audio signal respectively input from a first microphone and a second microphone arranged substantially symmetrically with respect to a specific speaker,
The power of the added signal obtained by adding the first audio signal and the second audio signal is compared with the power of the difference signal obtained by subtracting the second audio signal from the first audio signal, and the presence or absence of the utterance by the speaker is determined. Provided with an utterance state estimating means for estimating,
An audio processing device characterized by the above. An audio processing device for processing a first audio signal and a second audio signal respectively input from a first microphone and a second microphone arranged substantially symmetrically with respect to a specific speaker,
Voice power calculation means for calculating the power of the added signal obtained by adding the first voice signal and the second voice signal to obtain voice power;
Noise power calculating means for calculating the power of the difference signal obtained by subtracting the second audio signal from the first audio signal to obtain noise power;
A difference between the voice power and the noise power, and a speech state estimating means for comparing the noise power to estimate the presence or absence of speech;
An audio processing device comprising:

Images