JP2019219556A - Voice recognition support system - Google Patents

Abstract

To improve S/N ratio by removing a stationary noise in addition to a time-varying noise and increase a voice recognition rate of a target voice signal even when the level of the input target voice signal is low.SOLUTION: An input AGC processing unit 3, a subtractive beamforming processing unit 4, a time-varying noise spectrum estimation processing unit 6, a stationary noise spectrum estimation processing unit 7, a target voice signal extraction processing unit 9, a target voice section detection processing unit 10 and an output AGC processing unit 13 are included. The stationary noise spectrum estimation processing unit 7 operates in a noise section detected by the target voice section detection processing unit 10.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a voice recognition support system for removing a noise component from a received voice signal included in a received signal obtained by a plurality of microphones (hereinafter referred to as microphones) to extract a target voice signal.

  In order to extract a target audio signal having a high S / N ratio by removing noise components from a received audio signal included in a received signal obtained by the microphones using a plurality of microphones, an audio signal as shown in FIG. A recognition support system has been proposed (Non-Patent Document 1).

  In FIG. 10, 21L and 21R are microphones arranged at predetermined intervals. Reference numeral 22 denotes an A / D conversion processing unit, which performs A / D conversion of a received audio signal received by the microphones 21L and 21R to generate a frame signal in a predetermined time unit.

  Reference numeral 23 denotes a subtraction type beamforming processing unit which takes in the received audio signals received by the microphones 21L and 21R and calculates the received audio signals in the directions other than the specific direction as noise components. Reference numeral 24 denotes an FFT (Fast Fourier Transform) processing unit for converting the noise component obtained by the subtraction type beamforming processing unit 23 from the time domain to the frequency domain.

  Reference numeral 25 denotes a time-varying noise spectrum estimation processing unit which estimates the spectrum of a noise component that changes with time from the noise component output from the FFT processing unit 24. Reference numeral 26 denotes an FFT processing unit that converts a received audio signal output from the A / D conversion processing unit 22 from a time domain to a frequency domain.

  Reference numeral 27 denotes a target voice signal extraction processing unit which subtracts the spectrum of the noise component estimated by the time-varying noise spectrum estimation processing unit 25 from the spectrum of the received voice signal fetched from the FFT processing unit 26 to obtain the spectrum of the target voice signal. Is extracted.

  Reference numeral 28 denotes an IFFT (Inverse Fast Fourier Transform) processing unit that performs an inverse transform of the target audio signal extracted from the target audio signal extraction processing unit 27 from a frequency domain to a time domain signal.

  In this voice recognition support system, a target voice signal can be extracted by removing noise components from the voice signals received by the microphones 21L and 21R.

Noise and Noise Reduction Method Using Spectral Subtraction Using Microphone Pair, Mizumachi and Akagi, IEICE Transactions on Electronics, Information and Communication Engineers A Vol.J82-A, No.4, pp.503-512, April 1999

  However, in the speech recognition support system of FIG. 10, the time-varying noise spectrum estimated by the time-varying noise spectrum estimating processing unit 25 follows the noise component changing over time to extract the noise component spectrum from the spectrum of the received speech signal. Although it can be removed, the spectrum of stationary noise such as environmental noise and the spectrum of vehicle running noise cannot be estimated, the performance of removing stationary noise is low, and it is desired to improve the S / N ratio. I was Further, when the level of the target voice signal included in the received voice signal is low, the target voice signal is buried in a noise component, and the voice recognition rate of the target voice signal is low.

  An object of the present invention is to improve the S / N ratio by removing stationary noise in addition to time-varying noise, and to increase the speech recognition rate of the target voice signal even when the level of the input target voice signal is low. The object of the present invention is to provide a speech recognition support system which has been developed.

To achieve the above object, the invention according to claim 1 provides an input AGC processing unit for receiving a received signal and outputting a received audio signal adjusted to a predetermined signal level, and an output from the input AGC processing unit. A subtractive beamforming processing unit that captures the received voice signal and extracts a received voice signal in a direction other than a specific direction as a noise component; and a noise component that captures the noise component extracted by the subtractive beamforming processing unit and changes over time. A time-varying noise spectrum estimating section for estimating a spectrum, a stationary noise spectrum estimating section for taking in a noise component extracted by the subtraction type beamforming processing section and estimating a noise component spectrum that constantly occurs, and The received voice signal output from the AGC processing unit is fetched to estimate the time-variant noise spectrum. A target audio signal extraction processing unit that extracts a target audio signal by removing the time-varying noise spectrum estimated by the processing unit and the stationary noise component spectrum estimated by the stationary noise spectrum estimation processing unit; and the target audio signal extraction processing unit. An output AGC processing unit that takes in the target audio signal extracted in step (a) and adjusts the signal level of the target audio section; and outputs the target audio from the start timing and end timing of the target audio signal extracted by the target audio signal extraction processing unit. A target voice section detection processing section for detecting a section and detecting a section other than the target voice section as a noise section, wherein the stationary noise spectrum estimation processing section operates in the noise section detected by the target voice section detection processing section. It is characterized by doing.
According to a second aspect of the present invention, in the speech recognition support system according to the first aspect, the stationary noise spectrum estimation processing section accumulates a spectrum of a noise component detected in the noise section to generate a stationary noise spectrum. It is characterized by estimation.
According to a third aspect of the present invention, in the voice recognition support system according to the first or second aspect, the input AGC processing section outputs the level-compressed reception signal when the target voice section is longer than a first set time. When the noise section is longer than a second set time, level amplification is performed to output the received voice signal in a range where the level of the received voice signal does not exceed the first set value.
According to a fourth aspect of the present invention, in the voice recognition support system according to any one of the first to third aspects, the output AGC processing unit is configured to set a level of the target audio signal input to the output AGC processing unit to a second set value. The target audio signal is selectively level-amplified within a range not exceeding.
According to a fifth aspect of the present invention, in the voice recognition support system according to the first, second, third, or fourth aspect, a means for adjusting a start timing of the target voice signal is provided.

  According to the present invention, since the stationary noise spectrum estimation processing unit is provided and the stationary noise spectrum estimation processing unit is operated in the noise section, the time-varying noise estimation and the stationary noise estimation can be processed in parallel, and any noise Can be reduced, and the S / N ratio of the target audio signal can be greatly improved. Further, since the input AGC processing unit and the output AGC processing unit are provided, the speech recognition rate of the target audio signal can be increased even if the level of the input target audio signal is low.

It is a functional block diagram of a speech recognition support system of a 1st example. 5 is a flowchart of an input AGC processing unit. (A), (b) is explanatory drawing which shows the relationship between a microphone, target sound, and noise, (c) is a flowchart of a subtraction type beamforming processing part. It is a flowchart of a time-varying noise spectrum estimation process. It is a flowchart of a stationary noise spectrum estimation process. It is a flowchart of a target audio signal extraction process. It is a flowchart of a target voice section detection process. (A) is a characteristic diagram of entropy of a normal received voice signal, (b) is a characteristic diagram of entropy of an excessively large received voice signal, and (c) is a characteristic diagram of entropy of a too small received voice signal. It is a flowchart of an output AGC processing unit. It is a functional block diagram of the conventional voice recognition support system.

  FIG. 1 shows a speech recognition support system according to one embodiment of the present invention. Reference numerals 1L and 1R denote L-channel and R-channel microphones arranged at predetermined intervals. Reference numeral 2 denotes an A / D conversion processing unit, which generates a frame signal in a predetermined time unit by performing A / D conversion on a reception signal received by the microphones 1L and 1R. Reference numeral 3 denotes an input AGC processing unit that adjusts the level of a received signal extracted from the A / D conversion processing unit 2 and outputs a received audio signal.

  Reference numeral 4 denotes a subtraction type beamforming processing unit which takes in two received audio signals output from the input AGC processing unit 3 and calculates a received audio signal in a direction other than a specific direction as a noise component. An FFT (Fast Fourier Transform) processing unit 5 converts the noise component obtained by the subtraction type beamforming processing unit 4 from the time domain to the frequency domain.

  Reference numeral 6 denotes a time-varying noise spectrum estimating processing unit which estimates a time-varying noise spectrum (frequency and level) that changes with time from a noise component taken in from the FFT processing unit 5. Reference numeral 7 denotes a stationary noise spectrum estimating processing unit, which estimates a stationary noise spectrum constantly generated by the noise component output from the FFT processing unit 5.

  Reference numeral 8 denotes an FFT processing unit that converts a received audio signal output from the input AGC processing unit 3 from a time domain to a frequency domain. Reference numeral 9 denotes a target voice signal extraction processing unit, which takes in the time-varying noise spectrum obtained by the time-varying noise spectrum estimation processing unit 6 and the stationary noise spectrum obtained by the stationary noise spectrum estimation processing unit 7 and performs FFT processing unit 8 Then, the spectrum of the target voice signal is extracted by removing the spectrum of the time-varying noise and the spectrum of the stationary noise from the spectrum of the received voice signal taken in from.

  Reference numeral 10 denotes a target voice section detection processing unit which captures the spectrum of the target voice signal obtained by the target voice signal extraction processing unit 9 and detects a boundary between the target voice section and the noise section. Reference numeral 11 denotes an IFFT (Inverse Fast Fourier Transform) processing unit for inversely converting the target audio signal output from the target audio signal extraction processing unit 9 from a frequency domain to a time domain signal.

  Reference numeral 12 denotes a delay processing unit that corrects an error in timing of switching from the noise section to the target voice section when the target voice section is detected by the target voice section detection processing unit 10. Reference numeral 13 denotes an output AGC processing unit that adjusts the level of the target audio signal subjected to the delay correction by the delay processing unit 12.

  Hereinafter, each processing unit will be described. FIG. 2 shows a processing flowchart of the input AGC processing unit 3. After the A / D conversion processing (S1), signals other than the audio band included in the received signal are removed by a high-pass filter and a low-pass filter, and a received audio signal is extracted (S2). If the level of the received audio signal exceeds the set value A, level compression is performed (S3, S4).

  When the target voice continuation detection time (target voice section) by the target voice section detection processing unit 10 exceeds the set time T1, level compression is performed (S5, S6). However, if not, it is next determined whether or not the continuous noise detection time (noise section) exceeds the set time T2 (S7). If the continuous noise detection time exceeds the set time T2, the level of the target audio signal is estimated to be low, and the level is amplified (S8). If the continuous noise detection time does not exceed the set time T2, the process is left as it is (S9). Then, the level of the received audio signal is changed according to steps S4, S6, and S8 (S10).

  With the above processing, when the target voice section is longer than the set time T1, the level of the received voice signal is compressed, and when the noise section is longer than the set time T2, the level of the received voice signal does not exceed the set value A. Signal level amplification is performed.

FIGS. 3A and 3B are explanatory diagrams of the processing of the subtraction type beamforming processing unit 4, and FIG. 3B is a flowchart thereof. Assuming that the microphone 1L and the microphone 1R are arranged at a distance L1 as shown in FIG. 3A and the target sound indicated by a solid line and the noise indicated by a broken line are received by both the microphones 1L and 1R, the target sound is The time to reach the microphone 1L with respect to the microphone 1L is delayed by d, and the noise is detected by delaying the time to reach the microphone 1R with respect to the microphone 1L by τ (S11). Then, the noise component glr input to the microphone 1L and the noise component grl input to the microphone 1R are extracted by performing the calculation shown in the following equation (1) using the detected delay times d and τ (S12). These noise components glr and grl are output as noise components. 1 is a received audio signal of the microphone 1L, and r is a received audio signal of the microphone 1R.

  FIG. 4 is a flowchart of the processing of the time-varying noise spectrum estimation processing unit 6. After performing the FFT processing (S21) on the noise components glr and grl obtained by the subtraction type beamforming processing unit 4, time-varying noise spectrum estimation (S22) is performed.

The equation for calculating the time-varying noise spectrum estimation value is as shown in the following equation (2). N (ω) with ∧ indicates an estimated noise component in the frequency domain. G (ω) is the spectrum of the noise component immediately after the subtractive beamforming output is converted from the time domain to the frequency domain, and ε exceeds 0 and is a value sufficiently smaller than 1.

  FIG. 5 is a flowchart of the processing of the stationary noise spectrum estimation processing section 7. In the stationary noise spectrum estimation, while the time-varying noise spectrum estimation is an estimation of a noise component that changes in real time, a noise section in which a stationary noise component is detected by the target speech section detection processing unit 10 (subtraction) This is a process for improving the noise removal performance by detecting the residual target speech component included in the estimated noise that cannot be completely processed by the beamforming. The time-varying noise spectrum estimation result is delayed (S31), and when the target voice section detection processing result indicates a noise section (S32), a stationary noise spectrum estimation value is calculated (S33).

  When detecting the boundary between the noise section and the target voice section by the target voice section detection processing, the delay processing S31 is performed before the timing of switching to the voice section in order to correct the timing error of switching from the noise section to the target voice section. This is a process for calculating a stationary noise spectrum from the noise spectrum of the above to prevent the residual target speech component from being included.

The formula for calculating the stationary noise spectrum estimated value is the following accumulation formula. α is a coefficient for averaging (0 ≦ α ≦ 1). (N) represents the current frame, and (n-1) represents the previous frame. The second term on the right is the accumulated value of the stationary noise spectrum estimation values up to the current frame.

  FIG. 6 is a processing flowchart of the target audio signal extraction processing section 9. The result obtained by processing the processing result of the input AGC processing unit 3 from the time domain to the frequency domain signal by the FFT processing unit 8 (S41), the result estimated by the time-varying noise spectrum estimation processing unit 6, and the stationary noise spectrum estimation processing unit 7, the target audio signal extraction processor 9 calculates the spectrum of the target audio signal (S42).

The calculation formula of the estimated value of the spectrum of the target audio signal is the following formula (4). S (ω) with ∧ indicates a target audio signal in the frequency domain, and X (ω) indicates a received audio signal (including the target audio signal and noise component) in the frequency domain taken in from the FFT processing unit 8. β and γ are coefficients (0 ≦ β ≦ 1, 0 ≦ γ ≦ 1).

  FIG. 7 is a flowchart of the processing of the target voice section detection processing unit 10. Here, the target voice section and the noise section are determined from the target voice signal and the received voice signal containing the noise component. As shown in FIG. 8A, in the normal case, when the entropy (power) of the received audio signal exceeds the threshold h, the continuous period in which the entropy (power) exceeds the threshold h is the target audio section, and the continuous The period is a noise section. The target voice section is a section from the start timing ta of the target voice signal to the end timing tb of the target voice signal, and the other sections are noise sections. The target voice section detection processing section 10 detects these two adjacent timings ta and tb, and controls the input AGC processing section 3, the delay processing section 12, and the output AGC processing section 13.

  First, the target audio signal extraction result obtained by the target audio signal extraction processing unit 9 is fetched and its entropy is calculated (S51). When the entropy becomes larger than the threshold value h, it is determined that the target audio signal has been detected (timing ta), and a hold time is set (S52, S53). This hold time is used to mask noise detection during the hold time even if the target audio signal is no longer detected after the start timing ta of the target audio signal once detected, that is, even if noise is detected. It is. In this manner, until the end of the target audio signal from the start timing ta of the detected target audio signal to the end of the hold time, even if the end of the target audio signal is detected, the end is ignored and the sound quality deterioration is prevented. This hold time is set to, for example, about 100 msec to 200 msec.

  If the target voice continuation detection time (target voice section) from when the start timing ta of the target voice signal is detected to when the end timing tb of the target voice signal is detected exceeds the set time T1, the input AGC processing unit 3 To perform level compression (S54, S55). As shown in FIG. 8B, when the overall entropy of the received voice signal is high, the noise component in addition to the target voice signal also exceeds the threshold value h, and all of them are erroneously recognized as the target voice signal. Therefore, when the target voice continuation detection time exceeds the set time T1, the level of the received voice signal is compressed by the input AGC processing unit 3 so that the target voice signal and the noise component of the received voice signal can be identified.

  When the target sound continuous detection time is shorter than the set time T1, the start timing ta of the target sound signal is corrected (S56). This correction is for correcting the timing ta (correcting the timing ta before the actual timing ta) so that the target audio signal detection processing has a margin. The delay time for the correction is calculated for the target audio signal of the immediately preceding frame (S57), and the delay time is set in the delay processing unit 12.

  On the other hand, if the entropy is less than the threshold value h, it is determined in step S52 that the target audio signal is no longer detected, and the control waits until the hold time set in step S53 expires (S58). When the hold time expires, the continuous detection period of noise section (noise section) from the detection of the target voice section end timing tb to the detection of the next target voice section start timing ta exceeds the set time T2. If there is, level amplification is performed in the input AGC processing unit 3 described in FIG. 2 (S59, S60). As shown in FIG. 8 (c), if the entropy of the received voice signal is low overall and does not reach the threshold h even if the target voice signal is included, the whole is erroneously recognized as a noise component as it is. Therefore, the level of the received voice signal is amplified so that the target voice signal and the noise component can be distinguished. If the noise section continuous detection time does not exceed the set time T2, the level is not changed in the input AGC processing unit 3 (S61).

  FIG. 9 is a processing flowchart of the delay processing unit 12 and the output AGC processing unit 13. The target voice signal output from the target voice signal extraction processing unit 9 and restored from the frequency domain to the time domain signal by the IFFT processing unit 11 is converted from the noise section detected by the target voice section detection processing unit 10 to the target voice section. The error of the switching timing ta is corrected by the delay processing in the delay processing unit 12 (S71). This delay processing is performed to match the processing in the output AGC processing unit 13.

  If the level of the target audio signal subjected to the delay processing exceeds the set value B, the output AGC processing unit 12 performs level compression (S72, S73, S74). When the target voice section is detected by the target voice section detection processing section 10, the level is amplified by the output AGC processing section 13 (S75, S76). (S75, S77). In this way, the output AGC processing unit 13 selectively amplifies the level of the target audio signal within a range where the level of the input target audio signal does not exceed the set value B.

  As described above, according to the voice recognition support system of the present embodiment, the target voice input to the microphones 1L and 1R from the direction indicated by the solid line in FIG. 3 and the noise input to the microphones 1L and 1R from the direction indicated by the broken line in FIG. The target voice can be separated and taken out, and the target voice can be adjusted to a predetermined level by increasing the S / N ratio, which is suitable for voice recognition support.

  1L, 1R: microphone, 2: A / D conversion processing section, 3: input AGC processing section, 4: subtraction beamforming processing section, 5: FFT processing section, 6: time-varying noise spectrum estimation processing section, 7: stationary Noise spectrum estimation processing section, 8: FFT processing section, 9: target voice signal extraction processing section, 10: target voice section detection processing section, 11: IFFT processing section, 12: delay processing section, 13: output AGC processing section

Claims (5)

An input AGC processing unit for receiving a received signal and outputting a received audio signal adjusted to a predetermined signal level;
A subtraction-type beamforming processing unit that takes in the received audio signal output from the input AGC processing unit and extracts a received audio signal in a direction other than a specific direction as a noise component;
A time-varying noise spectrum estimation processing unit that takes in the noise components extracted by the subtraction type beamforming processing unit and estimates a noise component spectrum that changes over time;
A steady-state noise spectrum estimation processing unit that captures the noise component extracted by the subtraction-type beamforming processing unit and estimates a noise component spectrum that constantly occurs.
It takes in the received voice signal output from the input AGC processing unit and removes the time-varying noise spectrum estimated by the time-varying noise spectrum estimation processing unit and the stationary noise component spectrum estimated by the stationary noise spectrum estimation processing unit. A target audio signal extraction processing unit for extracting a target audio signal by
An output AGC processing unit that takes in the target audio signal extracted by the target audio signal extraction processing unit and adjusts a signal level of a target audio section;
A target voice section detection processing section that detects the target voice section from the start timing and the end timing of the target voice signal extracted by the target voice signal extraction processing section and detects a section other than the target voice section as a noise section,
The speech recognition support system according to claim 1, wherein the stationary noise spectrum estimation processing section operates in the noise section detected by the target speech section detection processing section. The speech recognition support system according to claim 1,
The speech recognition support system according to claim 1, wherein the stationary noise spectrum estimation processing unit estimates a stationary noise spectrum by accumulating a spectrum of a noise component detected in the noise section. The speech recognition support system according to claim 1 or 2,
The input AGC processing unit outputs the level-compressed received signal when the target voice section is longer than a first set time, and amplifies the level of the received voice signal when the noise section is longer than a second set time. Output the received voice signal within a range not exceeding a first set value. The speech recognition support system according to claim 1, 2 or 3,
Speech recognition support, wherein the output AGC processor selectively amplifies the level of the target audio signal within a range where the level of the target audio signal input to the output AGC processor does not exceed a second set value. system. The speech recognition support system according to claim 1, 2, 3, or 4,
A speech recognition support system comprising means for adjusting a start timing of the target speech signal.

Images