JP4173462B2 - Microphone position determination method, microphone position determination device, microphone position determination program - Google Patents

Description

  The present invention relates to a microphone position determination method, a microphone position determination apparatus, and a microphone position determination program for determining an optimal microphone position when setting the position of a microphone used for speech recognition.

  Conventionally, in order to determine the optimum microphone position, the spatial transfer characteristics from the speaker to the microphone placed at the speaker position, that is, the impulse response and background noise mixed in the microphone in the usage environment are recorded, and the noise is clean. Each speech obtained by convolving the impulse response measured at each microphone position and superimposing the recorded noise on the close-talked clean speech recorded in an environment where the speaker's mouth and microphone are close to each other A voice recognition experiment was actually performed on the voice simulating the recorded voice at the microphone position, and the microphone position with the highest recognition performance was set as the optimum microphone position.

In the method of creating simulated speech and determining the optimal microphone position through actual recognition experiments, you can create close speech speech for evaluation, or create speech data that convolves the impulse response and superimposes the recorded noise. The time, cost, and data storage area for actual speech recognition experiments are required.
The present invention has been made in view of the above. The purpose of the present invention is not to create a simulated voice at each microphone position from the measured impulse response and recorded background noise, and to actually perform a recognition experiment. Another object of the present invention is to provide an apparatus capable of determining an optimum microphone position from impulse response and background noise data.

In order to achieve the above object, the present invention proposed in claim 1 is a microphone position determination method for determining an optimum microphone position used for speech recognition. In the microphone position determination method, a microphone position is moved sequentially or a plurality of microphones are switched. The gist is to compare the ratio of the direct wave level of the impulse response at each microphone position and the reflected wave level, and to determine the microphone position having the largest ratio as the optimum microphone position.
In the present invention proposed in claim 1, the impulse response at each microphone position is measured. As shown in FIG. 1, from the measured impulse response data, the maximum value of the absolute value level of the amplitude of the measured impulse response, or the sum of the absolute value levels of the time around the time of the maximum value is defined as D (direct wave level). . From the measured impulse response data, the sum of absolute value levels at times other than the direct wave level is R (reflected wave level) as shown in FIG. The microphone position having the largest D / R ratio is set as the optimum microphone position.

According to a second aspect of the present invention, there is provided a microphone position determining method for determining an optimum microphone position used for speech recognition, wherein the microphone position is sequentially moved or a plurality of microphones are switched to change the amplitude of an impulse response at each microphone position. The sum of the absolute value level and the ratio of the level of the section exceeding the width of the analysis frame used for speech recognition from the impulse response data is compared, and the microphone position with the largest ratio is set as the optimum microphone position. To do.
According to the present invention, the impulse response at each microphone position is measured. From the measured impulse response data, the maximum value of the absolute value level of the amplitude of the measured impulse response as shown in FIG. 2, or the sum of the absolute value levels of the time around the time of the maximum value is D (direct wave level). . As shown in FIG. 2, from the measured impulse response data, the sum of the absolute value levels of the amplitudes from the time of the direct wave to the analysis frame used for speech recognition from the impulse response data is defined as R (reflected wave level). . The microphone position having the largest D / R ratio is set as the optimum microphone position. For example, when recognition is performed using a characteristic parameter related to a logarithmic spectrum, such as a cepstrum or MFCC, which is often used for speech recognition, transmission that falls within the frame ignored in claim 2 by subtracting the long-time average. It is possible to reduce the influence of the characteristic (impulse response).

According to a third aspect of the present invention, there is provided a microphone position determining method for determining an optimum microphone position used for speech recognition, wherein the ratio of the direct wave level and the reflection level of the amplitude of the impulse response according to the first and second aspects. Based on the above, in addition to the determination that is optimal as the ratio is large, the determination that is optimal is performed as the background noise level mixed in each microphone position is low, and the best microphone position is determined with the best result of both determinations. This is the gist.
In the present invention proposed in claim 3, the impulse response at each microphone position is measured. The maximum value of the absolute value level of the amplitude of the measured impulse response or the sum of the absolute value levels of the time around the time of the maximum value is defined as D (direct wave level). Let R be the reflection level determined by the method of claims 1 and 2 described above. Let N (background noise level) be the average level of the absolute value of the amplitude of noise recorded at each microphone position. Then, the microphone position at which the sum of the ratio D / R between the direct wave level and the reflected wave level and the ratio D / N between the direct wave level and the background noise level is maximized is determined as the optimum microphone position. The magnitude of the direct wave level indicates the power level of the voice input to the microphone, and the ratio D / N between the direct wave level and the background noise level is S / N, which is the ratio of the voice power level to the background noise power level. Correlation is high. Since the magnitude of S / N greatly affects the performance of speech recognition, the recognition performance can be estimated by knowing the D / N.

  According to the present invention, recording of background noise at each microphone position, measurement of impulse response, and convolution of the measured impulse response without the need for voice recognition experiments from the creation of a simulated voice superimposed with background noise. The microphone position can be determined, and the processing time and calculation cost can be reduced.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 3 is a block diagram of the optimum microphone position determination apparatus proposed in claims 1 and 2 of the present invention. The optimum microphone position determination apparatus shown in FIG. 3 determines the optimum microphone position by comparing the level R of the reflected wave obtained from the impulse response measured at each microphone position with the level D of the direct wave. Each microphone position is evaluated by an evaluation function (for example, Expression (1)), and an optimum microphone position is determined.
Q = D / R (1)
Specifically, as shown in FIG. 3, the optimum microphone position determination apparatus of the present embodiment includes a recording gain adjustment module 100, an impulse response measurement module 200, an evaluation function calculation module 300, and an optimum microphone position determination module 400. It is comprised by.

  In the optimum microphone position determination apparatus configured as described above, the recording gain is first adjusted as an initial setting. In the recording gain adjustment module 100 of FIG. 3, as shown in FIG. 4, the reference signal is taken out from the reference signal memory 101, the gain is adjusted by the recording gain adjustment unit 102, and converted into an analog signal by the D / A conversion unit 103. , Reproduced from the speaker simulation speaker SP, recorded by the microphone M for voice recognition, the recorded signal is converted into a digital signal by the A / D converter 104, and the power level of the recorded signal is calculated by the recorded signal power level calculator 105 Then, it is determined whether or not the level of the signal recorded by the recording signal power level determination unit 106 is within an appropriate range. If the level is within the appropriate range, the level determination switch 107 is set to the appropriate level terminal 108 side, and the reference signal The power level of the recorded signal is stored in the power level memory 110. If it is out of the proper range, the level judgment switch 107 is set to the inappropriate level terminal 109 side, the recording gain is adjusted by the recording gain adjusting unit 102, and the reference signal is reproduced and recorded again. The reference signal used here may be one sentence or a plurality of voices, and the reproduction level from the speaker simulation speaker SP is set to be equal to the volume level of a normal speaker. This completes the initial setting of the recording gain.

  After the initial setting of the recording gain adjustment unit 102 is completed, the impulse response is measured by the impulse response measurement module 200 of FIG. In the impulse response measurement module 200 of FIG. 3, as shown in FIG. 5, the impulse response measurement signal stored in the impulse response measurement signal memory 201 is adjusted by the initial recording gain adjustment unit 102, and the D / D The signal is converted into an analog signal by the A conversion unit 103 and reproduced from the speaker simulation speaker SP. At this time, the gain of the impulse response measurement signal stored in the impulse response measurement signal memory 201 is the same as the gain of the reference signal stored in the reference signal memory 101.

Therefore, the sound pressure of the impulse response measurement signal emitted from the speaker simulation speaker SP is the same as the sound pressure of the reference signal read from the reference signal memory 101.
The impulse response measurement signal reproduced from the speaker simulation speaker SP is recorded by the voice recognition microphone M, converted into a digital signal by the A / D converter 202, converted into an impulse response by the impulse response calculator 203, Stored in the impulse response memory 204.
Then, the evaluation function calculation module 300 in FIG. 3 calculates an evaluation function value indicating the optimum degree of microphone position from the measured impulse response. In the evaluation function calculation module 300 of FIG. 3, as shown in FIG. 6, the direct wave level calculation means 301 calculates the direct wave level D from the measured value of the impulse response read from the impulse response memory 204 in which the measured impulse response is stored. The reflected wave level R is calculated by the reflected wave level calculation means 302.

Here, in the microphone position determination method and the microphone position determination device proposed in claims 1 and 4 of the present invention, the direct wave level is the time wave of the maximum value of the amplitude of the impulse response shown in FIG. Is calculated as a direct wave, and a reflected wave level R is calculated using a wave at a time other than the direct wave as a reflected wave.
On the other hand, the microphone position determination method and the microphone position determination device proposed in claims 2 and 5 of the present invention employ the same calculation method as in claims 1 and 4 with respect to the direct wave level D, but with respect to the reflected wave level R. As shown in FIG. 2, the reflection level R is calculated by using the reflected wave as the reflected wave after the time exceeding the analysis frame width (several tens of ms) used for speech recognition from the time indicating the direct wave of the impulse response.

Based on the calculated direct wave level D and reflected wave level R, the evaluation function calculation unit 303 calculates an evaluation function using, for example, Expression (1), and inputs the calculation result to the optimum microphone position determination module 400.
As shown in FIG. 6, the optimal microphone position determination module 400 includes a maximum value determination switch 401, an optimal microphone position candidate memory 404, an evaluation function maximum value memory 405, and a non-optimal microphone position candidate memory 406.
The maximum value determination switch 401 compares the obtained evaluation function value with the function value recorded in the evaluation function maximum value memory 405, and determines whether or not the evaluation function value is larger than the evaluation function value measured at another microphone position. If it is larger, the switch is set to the maximum value terminal 402 side, the maximum value of the evaluation function is updated, the value stored in the evaluation function maximum value memory 405 is rewritten, the optimal microphone position is updated, and the optimal microphone position candidate The value in the memory 404 is rewritten with the ID (number representing the microphone position) of the microphone position.

When the evaluation function value is smaller than the evaluation function value measured at another microphone position, the switch 401 is set to the non-maximum value terminal 403 side, and the ID of the microphone position is added to the non-optimum microphone position candidate memory 406. At the stage where evaluation has been completed for all microphone positions, the optimal microphone position is determined based on the microphone position ID stored in the optimal microphone position candidate memory 404.
FIG. 7 shows the overall configuration of a microphone position determining apparatus for realizing the microphone position determining method proposed in claims 3 and 6 of the present invention. A configuration different from the microphone position determination apparatus shown in FIG. 3 is that a noise level ratio adjustment module 50 and a noise recording module 60 are added between the recording gain adjustment module 100 and the impulse response measurement module 200.

Based on the ratio of the direct wave level D and the reflected wave level R described in FIGS. 1 to 6 with this added configuration, in addition to the determination that the larger the ratio is, the background noise mixed at each microphone position is reduced. The determination that is optimal as the level N is lower is added to obtain a determination including the influence of background noise.
The evaluation function in this case is calculated by the following equation.
Q = D / R + k · D / N (2)
k: Correction coefficient for gain difference between impulse response and recording level The recording gain adjustment module 100 and impulse response measurement module 200 shown in FIG. 7 are the same as those in FIGS. The part will be described.

In the noise level ratio adjustment module 50 shown in FIG. 7, as shown in FIG. 8, nothing is reproduced from the speaker simulation speaker SP, the background noise is recorded by the voice recognition microphone M, and the recorded signal is A / The digital signal is converted by the D converter, the noise power level is calculated by the background noise level measuring means 52, stored in the background noise power level memory 53, and the reference signal power level memory stored in the recording gain adjustment module 100 of FIG. The reference signal power level stored in 110 is stored in the reference / noise level ratio memory 55.
After completion of the initial setting, noise is recorded by the noise recording module 60 shown in FIG. 9 at each microphone position. In the noise recording module 60, as shown in FIG. 9, the background noise signal recorded by the speech recognition microphone M is converted into a digital signal by the A / D converter 61, and the noise power level is calculated by the noise power level calculator 62. Is calculated and stored in the noise power level memory 63.

  In this embodiment, the evaluation function calculation module 300 reads the impulse measurement value from the impulse response memory 204 as shown in FIG. 10, and the direct wave level calculation means 301 and the reflected wave level calculation means 302 perform the direct wave level D and the reflected wave level. R is calculated. Further, the background noise power level is read from the noise power level memory 63, and the background noise power level is adjusted by the noise gain adjustment unit 304 based on the ratio of the reference level / noise level stored in the reference / noise level ratio memory 55. Then, the direct wave level and the reflected wave level are matched with the range. That is, the correction coefficient k shown in Expression (2) is determined. From the obtained direct wave level D, reflected wave level R, and noise level N, an evaluation function such as Equation (2) is calculated by the evaluation function calculation unit 303, for example. The calculation result of the evaluation function calculation unit 303 is sorted by the optimum microphone position determination module 400, and the optimum microphone position is determined.

Below, the outline | summary of the microphone position determination program of this invention is demonstrated with reference to the flowchart shown in FIG.
In what state it is determined whether the system has been initialized (step S91).
If the initial setting is not performed, the recording gain is adjusted (step S92), and the noise level ratio is adjusted (step S93).
When the initial setting is completed, the following procedure is repeated until the investigation of all microphone positions is completed (step S94).

First, noise is recorded and the power level of the noise is obtained (step S95).
Next, the impulse response is measured (step S96).
A direct wave level, a reflected wave level, and a noise level are obtained from the obtained impulse response and the recorded noise power level, and an evaluation function for evaluating the optimum microphone position is calculated (step S97).
It is determined whether or not the evaluation function value is maximum as compared with the evaluation function values of other microphones (step S98).

If it is not the maximum, the microphone position is registered in the list of non-optimal microphone positions (step S99), and the process returns to the determination of whether there is another microphone position (step S94).
If it is the maximum, the maximum evaluation function value is updated (step S100), the microphone position is replaced with the optimum microphone position candidate (step S101), and the process returns to the determination of whether there is another microphone position (step S94).
When all microphone positions have been determined, the optimum microphone position is output from the ID stored in the optimum microphone position candidate memory (step S102).

To exit.
The microphone position determination apparatus of the present invention described above can be realized by causing a computer to decode a microphone position determination program. The microphone position determination program proposed in the present invention is written in a computer-readable program language, recorded on a recording medium such as a magnetic disk or CD-ROM, and installed in the computer from the recording medium, or a communication line. Installed in the computer, and decoded by a central processing unit provided in the computer to function as a microphone position determination device.

  The microphone position determining apparatus according to the present invention is utilized when determining a microphone position for taking in voice for voice recognition when a device having a voice recognition function such as an automatic voice guidance device is installed.

The figure for demonstrating the calculation method of the direct wave level D and the reflected wave level R from the impulse response shown in FIG. The figure for demonstrating the calculation method of the direct wave level D and the reflected wave level R from the impulse response shown in Claim 2. The block diagram for demonstrating embodiment of the microphone position determination apparatus by this invention. The block diagram for demonstrating the recording gain adjustment module of the microphone position determination apparatus shown in FIG. The block diagram for demonstrating the impulse response measurement module of the microphone position determination apparatus shown in FIG. The block diagram for demonstrating the evaluation function calculation module and optimal microphone position determination module of the microphone position determination apparatus shown in FIG. The block diagram for demonstrating other embodiment of the microphone position determination apparatus by this invention. The block diagram for demonstrating the noise level ratio adjustment module of the microphone position determination apparatus shown in FIG. The block diagram for demonstrating the noise recording module of the microphone position determination apparatus shown in FIG. The block diagram for demonstrating the evaluation function calculation module and optimal microphone position determination module of the microphone position determination apparatus shown in FIG. The flowchart for demonstrating the outline | summary of the microphone position determination program which makes a computer function the microphone position determination apparatus shown in FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 50 Noise level ratio adjustment module 203 Impulse response calculation part 60 Noise recording module 204 Impulse response memory 100 Recording gain adjustment module 301 Direct wave level calculation means 200 Impulse response measurement module 302 Reflected wave level calculation means 300 Evaluation function calculation module 303 Evaluation function calculation 400 Optimum microphone position determination module 401 Maximum value determination switch SP Speaker simulation speaker 404 Optimal microphone position candidate memory M Speech recognition microphone 405 Evaluation function maximum value memory 101 Reference signal memory 406 Non-optimal microphone position candidate memory 102 Recording gain adjustment unit 51 A / D converter 103 D / A converter 52 Background noise level measuring means 104 A / D converter 53 Background noise power level memory 105 Recorded signal power level calculator 54 Reference / Noise Level Ratio Calculation Unit 106 Recording Signal Power Level Determination Unit 55 Reference / Noise Level Ratio Memory 107 Level Determination Switch 61 A / D Conversion Unit 110 Reference Signal Power Level Memory 62 Noise Power Level Calculation Unit 201 Impulse Response Measurement Signal Memory 63 Noise power level memory 202 A / D converter

Claims (7)

  The impulse response from the speaker simulation speaker to the voice recognition microphone is measured for each of a plurality of positions by changing the position of the voice recognition microphone, and the time wave of the maximum value of the amplitude of the impulse response measured for each position, Alternatively, a direct wave level is calculated using a wave at a time around it as a direct wave, a reflected wave level is calculated using a wave at a time other than the direct wave as a reflected wave, and the ratio between the direct wave level and the reflected wave level is maximum. The microphone position determination method which makes the microphone position which becomes the optimal microphone position.   The impulse response from the speaker simulation speaker to the voice recognition microphone is measured for each of a plurality of positions by changing the position of the voice recognition microphone, and the time wave of the maximum value of the amplitude of the impulse response measured for each position, Or calculate the direct wave level as a direct wave of the surrounding time wave, calculate the reflected wave level as a reflected wave after the time exceeding the analysis frame width used for speech recognition from the time showing the direct wave of the impulse response, A microphone position determination method in which the microphone position at which the ratio of the direct wave level and the reflected wave level is maximum is the optimum microphone position.   3. The microphone position determination method according to claim 1, wherein a background noise level mixed in the voice recognition microphone is measured for each of a plurality of positions by changing a position of the voice recognition microphone, and the direct wave level is measured. And a background noise level ratio, and a microphone position determination method in which the microphone position at which the sum of the ratio and the ratio of the direct wave level and the reflected wave level is maximized is the optimum microphone position. A device for determining the optimal microphone position used for speech recognition,
A speaker simulation speaker simulating a speaker placed at an assumed speaker's speaking position, a voice recognition microphone, and an impulse response from the speaker simulation speaker to the voice recognition microphone is changed in position of the voice recognition microphone. Impulse response measurement means that measures at multiple positions, and calculates the direct wave level using the wave of the maximum amplitude of the impulse response measured at each position or the surrounding time as a direct wave A direct wave level calculating means, a reflected wave level calculating means for calculating a reflected wave level using a wave at a time other than the direct wave as a reflected wave, and a microphone position at which a ratio of the direct wave level and the reflected wave level is maximum. A microphone position determining device, comprising: an optimum microphone position determining means for setting an optimal microphone position. A device for determining the optimal microphone position used for speech recognition,
A speaker simulation speaker simulating a speaker placed at an assumed speaker's speaking position, a speech recognition microphone, and an impulse response from the previous speaker simulation speaker to the speech recognition microphone, changing the position of the speech recognition microphone And an impulse response measuring means for measuring at each of a plurality of positions, and directly calculating a direct wave level by using a wave at the time of the maximum amplitude of the impulse response measured at each position or a wave at a time around it as a direct wave. A wave level calculating means, a reflected wave level calculating means for calculating a reflected wave level from a time indicating a direct wave of an impulse response as a reflected wave after a time exceeding an analysis frame width used for speech recognition, the direct wave level and the reflected wave What is claimed is: 1. A microphone position determining apparatus comprising: an optimum microphone position determining unit that sets a microphone position having a maximum wave level ratio as an optimal microphone position.   6. The microphone position determining apparatus according to claim 4, wherein the background noise level is recorded by changing the position of the voice recognition microphone and recording background noise mixed in the voice recognition microphone for each of a plurality of positions. A background noise level measurement means for measuring the direct wave level and the background noise level, and a microphone position at which the sum of the ratio and the ratio of the direct wave level and the reflected wave level is maximized is determined as an optimum microphone position. A microphone position determining apparatus comprising: an optimum microphone position determining means.   A microphone position determination program, which is written in a computer-readable program language, and causes the computer to execute any function of the microphone position determination apparatus according to any one of claims 4 to 6.

Images