JP2020134566A - Voice processing system, voice processing device and voice processing method - Google Patents

Abstract

To provide a voice processing system capable of improving accuracy of suppressing a noise component of a voice signal even when a number or positions of noise sources change.SOLUTION: The voice processing system includes: a first voice acquisition unit which acquires a voice signal including a first voice component and a voice component other than the first voice component; a plurality of second voice acquisition units which acquire a plurality of reference signals including a second voice component and a voice component other than the second voice component; a first adaptive filter which passes two or more reference signals among the plurality of reference signals and generates a first pass signal; a plurality of second adaptive filters which pass a different single reference signal among the plurality of reference signals and generate a plurality of second pass signals; and a control unit which determines an adaptive filter of a controlled object among the first adaptive filter and the plurality of second adaptive filters based on the voice signal, the first pass signal, and the plurality of second pass signals, and controls the filter coefficient of the adaptive filter of the controlled object.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to a voice processing system, a voice processing device, and a voice processing method.

An echo canceller suitable for use in an in-vehicle voice recognition device or a hands-free telephone is known (see Patent Document 1). This echo canceller switches the number of adaptive filters and the number of taps that operate in the echo canceling process according to the number of sound sources.

Further, a signal processing device that eliminates noise, interfering signals, echoes, etc. mixed in a signal is known (see Patent Document 2). This signal processing device adjusts the update amount of the corresponding adaptive filter according to the magnitude of the audio signal of each sound source in an environment where a plurality of sound sources exist.

Japanese Patent No. 4889810 Japanese Patent No. 6363324

However, in the echo canceller of Patent Document 1, since the positions of a plurality of sound sources are known, the number of adaptive filters and the number of taps can be adjusted, but the position of the speaker, which is a noise source generated in the vehicle interior, changes. Therefore, it is difficult to deal with it. Further, in the signal processing apparatus of Patent Document 2, when the number of sound sources which are noise sources changes, it takes some time for the adaptive filter to converge, and the sound quality during that time may deteriorate.

The present disclosure has been made in view of the above circumstances, and is a voice processing system, a voice processing device, and a voice capable of improving the suppression accuracy of the noise component of the voice signal to be acquired even when the number or position of the noise source changes. Provides a processing method.

One aspect of the present disclosure is a first audio acquisition unit that acquires an audio signal including an audio component other than the first audio component and the first audio component, and a second audio component other than the second audio component and the second audio component. A first adaptation in which a plurality of second audio acquisition units that acquire a plurality of reference signals including audio components and two or more reference signals among the plurality of reference signals are passed to generate a first pass signal. A filter, a plurality of second adaptive filters that pass a different single reference signal among the plurality of reference signals to generate a plurality of second pass signals, and the audio signal and the first pass signal. The adaptive filter to be controlled is determined from the first adaptive filter and the plurality of second adaptive filters based on the plurality of second passing signals, and the filter coefficient of the adaptive filter to be controlled is controlled. It is a voice processing system including a control unit for the operation.

One aspect of the present disclosure includes a control unit that acquires a voice signal including a first voice component and a voice component other than the first voice component, and a second voice component and a voice component other than the second voice component. A first that includes a plurality of adaptive filters that acquire a plurality of reference signals, and the plurality of adaptive filters pass two or more reference signals among the plurality of reference signals to generate a first pass signal. The control unit includes the adaptive filter of the above and a plurality of second adaptive filters that pass a different single reference signal among the plurality of reference signals to generate a plurality of second pass signals. Based on the voice signal, the first passing signal, and the plurality of second passing signals, the adaptive filter to be controlled is determined from the first adaptive filter and the plurality of second adaptive filters, and the adaptive filter to be controlled is determined. It is a voice processing device that controls the filter coefficient of the adaptive filter to be controlled.

One aspect of the present disclosure is a plurality of references that acquire an audio signal including a first audio component and an audio component other than the first audio component, and include a second audio component and an audio component other than the second audio component. A signal is acquired, two or more of the plurality of reference signals generate a first pass signal that has passed through the first adaptive filter, and a different single reference signal among the plurality of reference signals A plurality of second pass signals that have passed through a plurality of second adaptive filters that pass are generated, and the first pass signal is based on the audio signal, the first pass signal, and the plurality of second pass signals. This is an audio processing method for determining an adaptive filter to be controlled from among the adaptive filter of the above and the plurality of second adaptive filters, and controlling the filter coefficient of the adaptive filter to be controlled.

According to the present disclosure, it is possible to improve the suppression accuracy of the noise component of the audio signal to be acquired even when the number or position of the noise source changes.

The figure which shows an example of the schematic structure of the voice processing system in 1st Embodiment Block diagram showing the hardware configuration of the audio processing device Flowchart showing the operation procedure of the voice processing device The figure which shows the hardware configuration of the voice processing apparatus in 2nd Embodiment

Hereinafter, a voice processing system according to an embodiment in which the voice processing system, the voice processing device, and the voice processing method according to the present disclosure are specifically disclosed will be described in detail with reference to the drawings as appropriate. However, more detailed explanation than necessary may be omitted. For example, detailed explanations of already well-known matters and duplicate explanations for substantially the same configuration may be omitted. This is to avoid unnecessary redundancy of the following description and to facilitate the understanding of those skilled in the art. It should be noted that the accompanying drawings and the following description are provided for those skilled in the art to fully understand the present disclosure, and are not intended to limit the subject matter described in the claims.

(First Embodiment)
FIG. 1 is a diagram showing an example of a schematic configuration of the voice processing system 5 according to the first embodiment. The voice processing system 5 is mounted on the vehicle 10. In the vehicle interior of the vehicle 10, for example, a driver's seat, a passenger's seat, and left and right rear seats are provided. The voice processing system 5 includes a plurality of microphones MC1 to MC4 and a voice processing device 20. The output of the voice processing device 20 is input to the voice recognition engine 40. The voice recognition result by the voice recognition engine 40 can be input to, for example, the car navigation device 50 and used as an operation signal of the car navigation device 50. The number of seats and the number of microphones are not limited to this.

In front of the driver's seat (for example, the front right side of the dashboard), a microphone MC1 that picks up the voice spoken by the driver hm1 may be arranged. In front of the passenger seat (for example, the front left side of the dashboard), a microphone MC2 that collects the voice spoken by the occupant hm2 may be arranged. A microphone MC3 that picks up the sound spoken by the occupant hm3 sitting in the left rear seat may be arranged in the backrest portion of the passenger seat. A microphone MC4 that picks up the sound spoken by the occupant hm4 sitting in the right rear seat may be arranged on the backrest of the driver's seat. The arrangement positions of the microphones MC1 to MC4 are not limited to this.

The microphones MC1 to MC4 may be either a directional microphone or an omnidirectional microphone. As the microphones MC1 to MC4, a small MEMS (micro electro mechanical systems) microphone may be used, or an electret condenser microphone (ECM: Electret Condenser Microphone) may be used. The microphones MC1 to MC4 may be microphones capable of beamforming, for example, a microphone array capable of forming directivity in the direction of each seat and collecting sound in the directivity direction.

A car navigation device 50 may be arranged on the dashboard of the vehicle 10. The voice processing device 20 and the voice recognition engine 40 may be housed and arranged inside the dashboard and inside the seat. The voice processing device 20 may be provided for each microphone and for each seat. For example, the voice processing device 20 may be the voice processing devices 21, 22, 23, 24 corresponding to the microphones MC1, MC2, MC3, and MC4, respectively.

Although it is illustrated in FIG. 1 that the voice processing devices 21, 22, 23, and 24 are separately configured, they may be configured by one voice processing device 20. That is, the voice processing device 20 may be composed of one voice processing unit and provided in plurality, may be composed of a plurality of voice processing units and provided in one, or may be composed of a plurality of voice processing units. It may be provided in plurality. Therefore, the voice processing system 5 may be provided with one voice processing device 20 or may be provided with a plurality of voice processing devices 20. Each voice processing device 20 (21, 22, 23, 24) may be configured by different hardware or may be configured by one common hardware.

Each voice processing device 20 may be arranged in, for example, any seat in the vehicle interior. Each voice processing device 20 may be arranged in each seat corresponding to each microphone. Each voice processing device 20 may be arranged in a dashboard or the like.

The voice recognition engine 40 recognizes the voice included in the output signal from at least one voice processing device 20, and outputs the voice recognition result. The voice recognition engine 40 generates a voice recognition result and a signal based on the voice recognition result (for example, an operation signal of the car navigation device 50). The voice recognition engine 40 may be a device separate from the voice processing device 20, or may be an integrated device incorporated in the voice processing device 20.

The car navigation device 50 is an example of an output destination of the voice processing system 5. The car navigation device 50 inputs an operation signal output from the voice recognition engine 40 and performs an operation corresponding to the operation signal. For example, the car navigation device 50 displays map data on a display and performs navigation for guiding the course of the vehicle.

The output destination of the voice processing system 5 is not limited to the car navigation device 50, but may be an electronic device such as a panel meter, a television, or a mobile phone.

Further, although FIG. 1 shows a case where four people are in the vehicle, the number of people in the vehicle is not limited to this number. The maximum number of passengers is limited to the maximum number of passengers in the vehicle, and the number of passengers may be 4, 6, 9, or the like within this range.

FIG. 2 is a block diagram showing a hardware configuration of the voice processing device 21 as the voice processing device 20. The voice processing devices 21, 22, 23, and 24 all have the same configuration and function. Here, the voice processing device 21 will be mainly described. The voice processing device 21 targets the voice spoken by the driver hm1 sitting in the driver's seat (voice signal for acquisition), and outputs a voice signal in which the crosstalk component is suppressed from the voice signal of the voice picked up by the microphone MC1. Output as a signal.

The voice processing device 21 is adapted to a voice input unit 29 for inputting a voice signal of the voice picked up by the microphone MC1, a plurality of (for example, four) adaptive filters F2, F3, F4, F5, and an adder 27. It has a configuration including a filter control unit 28.

The voice input unit 29 inputs the voice signal of the voice picked up by the microphone MC1 arranged in front of the driver hm1. This voice signal is a signal including the voice of the driver hm1 (sound of the target component) and noise including the voice of an occupant other than the driver hm1 (sound of the crosstalk component).

The adaptive filter F2 includes a plurality of (for example, three) adaptive filters F2A, F2B, and F2C. The adaptive filters F2A, F2B, and F2C are picked up by the microphone MC2, the microphone MC3, and the microphone MC4 in order to suppress the crosstalk component other than the sound of the driver hm1 contained in the sound picked up by the microphone MC1. The audio signal of the audio is input as a reference signal, and the passing signal that has passed through the adaptive filters F2A, F2B, and F2C is extracted. The adaptive filter F2 adds and outputs the passing signals extracted by the adaptive filters F2A, F2B, and F2C. The adaptive filters F2A, F2B, F2C may be physically separated.

The adaptive filter F3 refers to the voice signal of the voice picked up by the microphone MC2 in order to suppress the crosstalk component other than the voice component of the driver hm1 included in the voice signal of the voice picked up by the microphone MC1. It is input as a signal, and a passing signal that has passed through the adaptive filter F3 is output.

The adaptive filter F4 refers to the voice signal of the voice picked up by the microphone MC3 in order to suppress the crosstalk component other than the voice component of the driver hm1 included in the voice signal of the voice picked up by the microphone MC1. It is input as a signal, and a passing signal that has passed through the adaptive filter F4 is output.

The adaptive filter F5 refers to the voice signal of the voice picked up by the microphone MC4 in order to suppress the crosstalk component other than the voice component of the driver hm1 included in the voice signal of the voice picked up by the microphone MC1. It is input as a signal, and a passing signal that has passed through the adaptive filter F5 is output.

Here, the outline of the operation of the adaptive filter will be described. The adaptive filter is a filter that minimizes the cost function defined by the root mean square of the error signal. Here, it is illustrated that an FIR (Finite impulse response) filter is used as the adaptive filter, but other adaptive filters may be used.

When the adaptive filter is used, the output signal of the voice processing device 21, that is, the subtraction signal e (n) is represented by, for example, the equation (1). Each delay block represented by the equation (1) is called a tap. The FIR filter adapts to various filter characteristics by changing the tap weight and the number of tap stages (tap length). The tap weight and the number of tap steps (tap length) are examples of filter coefficients.

ここで、ｎは時刻を表す。ｄ（ｎ）は取得したい（ターゲットの）音声信号である。ｘ（ｎ）は参照信号である。参照信号とは、ターゲットの音声信号以外の音声信号の１つである。ｗｉはフィルタ係数（タップの重み）である。Ｉはタップ長である。

Here, n represents a time. d (n) is the (target) audio signal to be acquired. x (n) is a reference signal. The reference signal is one of the audio signals other than the target audio signal. wi is a filter coefficient (tap weight). I is the tap length.

ここで、αは、フィルタ係数の補正係数であり、更新幅（更新量）に相当する。 Further, the update of the filter coefficient in the LMS (Least Mean Square) algorithm is represented by the equation (2).

Here, α is a correction coefficient of the filter coefficient and corresponds to an update width (update amount).

Although it has been illustrated that LMS is used as an algorithm for updating the filter coefficient, the present invention is not limited to this, and other algorithms (for example, ICA (Independent Component Analysis) and NLMS (Normalized Least Mean Square)) may be used. ..

The adder 27 subtracts (subtractively adds) the passing signal output from the adaptive filter F2 from the target voice signal output from the voice input unit 29, and outputs this subtracted signal as an error signal. The adder 27 subtracts the passing signal output from the adaptive filter F3 from the target audio signal output from the audio input unit 29, and outputs this subtracted signal as an error signal. The adder 27 subtracts the passing signal output from the adaptive filter F4 from the target audio signal output from the audio input unit 29, and outputs this subtracted signal as an error signal. The adder 27 subtracts the passing signal output from the adaptive filter F5 from the target voice signal output from the voice input unit 29, and outputs this subtracted signal as an error signal.

The adaptive filter control unit 28 selects an error signal having the lowest signal level from a plurality of (for example, four) subtraction signals (error signals) output from the adder 27, and outputs the error signal as an output signal. .. The output signal of the adaptive filter control unit 28 is input to the voice recognition engine 40. The voice recognition engine 40 is an example of an output destination of the adaptive filter control unit 28. The output destination of the adaptive filter control unit 28 may be a speaker or the like that emits sound. At this time, the adaptive filter control unit 28 may output an output signal to the mobile terminal via a wireless communication network or the like. The output signal output to the mobile terminal may be output as voice from a speaker or the like of the mobile terminal.

The adaptive filter control unit 28 selects an adaptive filter corresponding to the passing signal corresponding to the error signal having the minimum signal level of the error signal from a plurality of (for example, four) adaptive filters F2 to F5, and selects the adaptive filter corresponding to the pass signal. Updates the filter coefficient of the selected adaptive filter so that is close to the value 0. When the adaptive filter control unit 28 updates the filter coefficient of any of the adaptive filters F2 to F5, the filter coefficient of the corresponding adaptive filter among the three adaptive filters F2A, F2B, and F2C included in the adaptive filter F2 is combined. May be updated. For example, when updating the adaptive filter F4, the adaptive filter control unit 28 may update the adaptive filter F2B to which the same reference signal C is input.

Although the adaptive filter control unit 28 exemplifies the selection of the adaptive filter corresponding to the reference signal corresponding to the error signal having the minimum signal level of the error signal, the criteria other than the minimum signal level are used. Based on this, an adaptive filter may be selected. For example, the adaptive filter control unit 28 may select any adaptive filter in which the signal level of the error signal is equal to or less than the threshold value th1.

The adaptive filter control unit 28 realizes various functions of the adaptive filter control unit 28, for example, by executing a program in which a processor (not shown) is held in a memory (not shown).

In FIG. 2, the voice signal A is a signal of voice (mainly including a target component) picked up by the microphone MC1 in the driver's seat. The reference signal B is a signal of voice (sound including non-target components and noise) collected by the microphone MC2 in the passenger seat and input to the adaptive filters F2 and F3. The reference signal C is an audio signal that is picked up by the microphone MC3 in the left rear seat and input to the adaptive filters F2 and F4. The reference signal D is an audio signal that is picked up by the microphone MC4 in the passenger seat and input to the adaptive filters F2 and F5.

The passing signals B', C', and D'are signals obtained by passing the reference signals B, C, and D of the voice picked up by the microphones MC2, MC3, and MC4 through the adaptive filters F2A, F2B, and F2C, respectively. Of the sound picked up by the microphone MC1, the crosstalk components other than the target component correspond to noise.

For example, in the voice processing device 21, when there is no voice signal A from the target seat (here, the driver's seat) and there are reference signals B to D due to utterance from the other passenger seat and the rear seat, the sound is picked up by the microphone MC1. A cross-talk component (leakage component) is included in the sound signal of the sound. The voice processing device 21 may update the adaptive filter to minimize the error signal. In this case, since there is no utterance in the driver's seat, the ideal error signal is a silent signal. Further, when there is an utterance voice signal in the driver's seat, the utterance included in the voice signal A is basically faster than the leak sound included in the reference signals B to D, so that the voice processing device 21 Cannot cancel the utterance contained in the voice signal A by the adaptive filter (causal law). Therefore, the audio processing device 21 maximizes the crosstalk component in the audio signal A by updating the adaptive filter so as to minimize the error signal regardless of whether the target audio signal is included or not. Can be reduced.

The adder 27 outputs a subtraction signal obtained by subtracting the passing signal E'of the adaptive filter F2 from the audio signal A. The passing signal E'is a signal B'+ C'+ D'that is the sum of the passing signals of the adaptive filters F2A, F2B, and F2C. Further, the adder 27 outputs a subtraction signal obtained by subtracting the passing signal B'of the adaptive filter F3 from the audio signal A. Further, the adder 27 outputs a subtraction signal obtained by subtracting the passing signal C'of the adaptive filter F4 from the audio signal A. Further, the adder 27 outputs a subtraction signal obtained by subtracting the passing signal D'of the adaptive filter F5 from the audio signal A.

FIG. 3 is a flowchart showing the operation procedure of the voice processing device 21. The voice input unit 29 of the voice processing device 21 inputs a voice signal A (a voice of the driver hm1 as a target component and a signal including a cross talk component) picked up by the microphone MC1 arranged in the driver's seat ( S1). The voice signal A may include a voice component of the driver hm1 as a target component and a voice component of the occupants hm2 to hm4 as a crosstalk component. The voice processing device 21 acquires the reference signals B, C, and D picked up by the microphones MC2, MC3, and MC4, respectively (S2). For example, the reference signals B, C, and D may include a voice component of the occupant hm2 as a main component other than the target component, and a voice component of the driver hm1 and the occupant hm3 and hm4 other than the main component.

The voice processing device 21 uses the reference signals B, C, and D to generate a passing signal that has passed the adaptive filter (S3). The adaptive filter F2 passes the reference signal B through the adaptive filter F2A, passes the reference signal C through the adaptive filter F2B, passes the reference signal D through the adaptive filter F2C, and adds the signals after each passage to the passing signal E. 'Generate. The adaptive filter F3 passes the reference signal B to generate a passing signal B'. The adaptive filter F4 passes the reference signal C to generate a passing signal C'. The adaptive filter F5 passes the reference signal D to generate a passing signal D'.

The adder 27 subtracts each passing signal E', B', C', D'from the audio signal A, and subtracts each subtracted signal A-E', AB', AC', AD'. Generate (S4). The adaptive filter control unit 28 selects an output signal based on the subtraction signals A-E', AB', AC', and AD'. For example, the adaptive filter control unit 28 performs subtraction in which the ratio of the target component among the subtraction signals A-E', AB', AC', and AD'is the maximum, that is, the signal level is the minimum. A signal (error signal) is selected as an output signal (S5).

The adaptive filter control unit 28 updates the filter coefficient of the adaptive filter corresponding to this output signal (S6). After that, the output signal output from the adaptive filter control unit 28 reflects the updated filter coefficient.

The adaptive filter control unit 28 outputs an output signal to the voice recognition engine 40 (S7).

The voice recognition engine 40 recognizes the voice included in the output signal, and transmits a voice instruction based on the recognition result to the car navigation device 50 which is an example of the output destination. The car navigation device 50 executes operations such as a destination, a map, and a navigation route according to the received voice instruction. Although the output signal is output to the voice recognition engine 40 here, it may be output to another device (for example, a speaker). The speaker produces a voice corresponding to the output signal. After this, the voice processing device 21 returns to the processing of S1. The output signal from the adaptive filter control unit 28 may be output to another device connected by wire, or may be output to another device connected wirelessly.

As described above, in the voice processing system 5 of the first embodiment, there are a plurality of microphones, a plurality of noise sources, ambient noise, operating sounds, speakers, speakers (speaking occupants), and the like in the vehicle interior. In this case, different adaptive filters are used, corresponding to the number of noise sources. Further, in the voice processing system 5, the adaptive filter control unit 28 switches the adaptive filter to be updated according to one or more noise sources. Further, when the number or position of the noise source generated in the vehicle interior changes, the adaptive filter control unit 28 identifies the adaptive filter corresponding to the position of the noise source and updates the filter coefficient of the specified adaptive filter. The adaptive filter control unit 28 switches the adaptive filter itself to be updated when updating the update amount and tap length of the adaptive filter according to the signal level of the error signal. As a result, even if the number or position of the noise sources changes, the S / N ratio of the microphone signal (for example, the audio signal A) corresponding to each noise source is improved.

If the microphone installed in the vehicle interior is a microphone array, the microphone array forms directivity toward the noise source and collects the sound (beamforming) before the noise suppression process. Therefore, the S / N ratio of the audio signal input to each microphone may be improved. As a result, the voice processing system 5 can enhance the noise suppression processing in the subsequent stage. That is, the microphone can efficiently pick up the noise sound generated by the noise source as a reference signal, and the adaptive filter control unit can update the adaptive filter so that the error signal is minimized. The adaptive filter can effectively perform the suppression process that cancels the noise sound well from the target audio signal.

Further, in the present embodiment, the voice processing device 20 separately inputs the reference signal of the voice emitted by the N people for each speaker in the situation where it is assumed that there are N speakers (for example, 4 people). It is illustrated that an adaptive filter and an adaptive filter for inputting a reference signal of a voice emitted by N people collectively for N people are provided. In this case, the audio processing device 20 can efficiently reduce noise components such as crosstalk components other than the target component from the target voice signal by using as few adaptive filters as possible. Further, the voice processing device 20 may be provided with an adaptive filter for each number of N or less people. For example, an adaptive filter for 2 people who inputs a reference signal of the voice of 2 people out of 4 people, an adaptive filter for 3 people who inputs a reference signal of the voice of 3 people, and a reference signal of the voice of 4 people. An adaptive filter for 4 people may be provided to input. In this case, the voice processing device can select the optimum adaptive filter according to the actual position and number of speakers, update the filter coefficient, and further improve the suppression performance of the crosstalk component.

As described above, the voice processing system 5 obtains the voice signal A including the target component (an example of the first voice component) and the crosstalk component (an example of the voice component other than the first voice component). An example of the first voice acquisition unit) may be provided. The voice processing system 5 includes a plurality of reference signals B and C including a principal component other than the target component (an example of the second voice component) and a component other than the main component (an example of the voice component other than the second voice component). A plurality of microphones MC1, MC3, MC4 (an example of a second audio acquisition unit) for acquiring, D may be provided. The voice processing device 21 of the voice processing system 5 passes two or more reference signals among the plurality of reference signals B, C, and D, and generates a passing signal E'(an example of the first passing signal). F2 (an example of a first adaptive filter) may be provided. The voice processing device 21 passes a different single reference signal among the plurality of reference signals B, C, and D, and generates a plurality of pass signals B', C', D'(an example of the second pass signal). A plurality of adaptive filters F3, F4, F5 (an example of a second adaptive filter) may be provided. The voice processing device 21 is a control target of the adaptive filter F2 and the plurality of adaptive filters F3, F4, F5 based on the voice signal A, the passing signal E', and the plurality of passing signals B', C', and D'. An adaptive filter control unit 28 (an example of a control unit) that determines an adaptive filter and controls the filter coefficient of the adaptive filter to be controlled may be provided.

As a result, the voice processing device 21 of the voice processing system 5 has a single voice signal to be acquired, a passing signal in which a reference signal other than this voice signal has passed through an adaptive filter for a plurality of signals, and a single reference signal other than this voice signal. Based on the passing signal that has passed through the adaptive filter for the signal, the adaptive filter can be determined in consideration of the state of each signal, and the filter coefficient of this adaptive filter can be controlled. Therefore, the voice processing device 21 can maintain a suitable filter coefficient by updating the applied filter that is highly relevant due to a change in the position of the speaker (an example of a noise source) or a change in the number of speakers, for example. For an applied filter that is less relevant, past learning results are unnecessarily updated, and it is possible to suppress a decrease in the filter efficiency of the applied filter. Further, by determining the adaptive filter for a plurality of signals and the adaptive filter for a single signal as the adaptive filter to be controlled, the voice processing device 21 can be used as a second speaker, for example, when there is one speaker. The application filter of can be used to efficiently remove audio components other than the first audio component. Further, when there are a plurality of speakers, the voice processing device 21 can efficiently remove voice components other than the first voice component by using the first applied filter.

Further, the adder 27 (an example of the control unit) may subtract the passing signal B'from the audio signal A to generate the subtraction signal AE'(an example of the first subtraction signal). The adder 27 subtracts different passing signals B', C', and D'from the audio signal A, and a plurality of subtracted signals AB', AC', AD'(of the second subtraction signal). An example) may be generated. The adaptive filter control unit 28 may determine the adaptive filter to be controlled based on the signal levels of the subtraction signals AE'and the subtraction signals AB', AC', and AD'. In the control unit, the adder 27 and the adaptive filter control unit 28 may be configured as separate bodies, or the adaptive filter control unit 28 may be configured to include the function of the adder 27.

As a result, the audio processing device 21 determines an adaptive filter having high removal efficiency in consideration of the removal efficiency other than the target component according to the signal level of the subtraction signal, controls the filter coefficient of the adaptive filter, and controls this. It is possible not to control the filter coefficient of the adaptive filter other than. Therefore, the voice processing device 21 can improve the suppression accuracy of the noise component of the voice signal to be acquired even when the number and position of the noise sources change.

Further, the adaptive filter control unit 28 sets the subtraction signals A-E'and each subtraction signal based on the signal levels of the subtraction signals A-E'and the subtraction signals AB', AC', and AD'. Any one of AB', AC', and AD' may be determined as an output signal. The adaptive filter control unit 28 may determine the adaptive filter to be controlled by any one of the adaptive filter F2 and the plurality of adaptive filters F3, F4, and F5 corresponding to the output signal.

As a result, the voice processing device 21 controls the filter coefficients of all the adaptive filters by controlling the filter coefficients of the adaptive filters corresponding to the output signals having a small crosstalk component used in the subsequent processing of the voice processing device 21. The filter coefficient can be controlled efficiently without the need to perform. In addition, the controlled filter coefficient can be expected to further reduce the crosstalk component of the subtraction signal, which is likely to be selected as the output signal later.

Further, the adaptive filter control unit 28 may be a signal having the minimum signal level among the subtraction signals A-E'and the subtraction signals AB', AC', and AD'.

As a result, the voice processing device 21 can select an adaptive filter that can suppress the crosstalk component most as an adaptive filter for which the filter coefficient is controlled.

Further, the adaptive filter control unit 28 has a signal level of the threshold value th1 (an example of the first threshold value) among the subtraction signals A-E'and the plurality of subtraction signals AB', AC', and AD'. The signal may be as follows.

As a result, the voice processing device 21 can select an adaptive filter capable of suppressing the crosstalk component more than a desired reference as the adaptive filter to be controlled by the filter coefficient.

Further, the microphones MC2, MC3, and MC4 are directed in the direction of the occupant hm2 to hm4 (an example of a sound source) as a speaker who emits the crosstalk component of the audio signal A in order to acquire the reference signals B, C, and D, respectively. May have sex.

As a result, the voice processing device 21 has directivity in a specific direction, so that the microphones MC2, MC3, and MC4 increase, for example, the voice components emitted by a specific speaker, and the voice emitted by a speaker other than the specific speaker. It can be obtained with less components. Therefore, the voice processing device 21 can use the adaptive filters F3, F4, and F5 exclusively for a specific speaker located in a specific direction by reducing the leakage of voices other than the specific speaker. .. Therefore, the voice processing device 21 can reduce the fluctuation during learning of the adaptive filter, and can efficiently suppress the voice component of a specific speaker from the voice signal.

Further, a plurality of voice processing devices 20 (21 to 25) including an adaptive filter F2, a plurality of adaptive filters F3 to F5, and an adaptive filter control unit 28 may be provided. Each voice signal (each target component included in the voice signal) acquired by each adaptive filter control unit 28 in the plurality of voice processing devices 20 may be different. Further, the combination of each adaptive filter F2 and each reference signal (main component other than each target component included in the reference signal) acquired by each adaptive filter F3 to F5 in the plurality of audio processing devices 20 may be different.

For example, the voice signal acquired by the adaptive filter control unit 28 of the voice processing device 21 may be a voice signal picked up by the microphone MC1, and the target component thereof may be the voice of the occupant hm1. The voice signal acquired by the adaptive filter control unit 28 of the voice processing device 22 may be a voice signal picked up by the microphone MC2, and the target component thereof may be the voice of the occupant hm2. The voice signal acquired by the adaptive filter control unit 28 of the voice processing device 23 may be a voice signal picked up by the microphone MC3, and the target component thereof may be the voice of the occupant hm3. The voice signal acquired by the adaptive filter control unit 28 of the voice processing device 24 may be a voice signal picked up by the microphone MC4, and the target component thereof may be the voice of the occupant hm4.

For example, the reference signal acquired by each of the adaptive filters F2 to F5 of the voice processing device 21 may be a signal picked up by the microphones MC2, MC3, and MC4, and the combination of the main components other than the target component is the occupant hm2, hm3. The voice of hm4 may be used. The reference signal acquired by each of the adaptive filters F2 to F5 of the voice processing device 22 may be a signal picked up by the microphones MC3, MC4, and MC1, and the combination of the main components other than the target component is the occupant hm3, hm4 and the driver. The voice of hm1 may be used. The reference signal acquired by each of the adaptive filters F2 to F5 of the voice processing device 23 may be a signal picked up by the microphones MC4, MC1 and MC2, and the main components other than the target component are the occupant hm4, the driver hm1 and the occupant. The voice of hm2 may be used. The reference signal acquired by each of the adaptive filters F2 to F5 of the voice processing device 24 may be a signal picked up by the microphones MC1, MC2, and MC3, and the main components other than the target component are the driver hm1 and the occupant hm2 and hm3. Voice is fine.

That is, the voice processing system 5 may include N voice processing devices 20 including N microphones (N is a natural number), N adaptive filters, and an adaptive filter control unit 28, respectively. The m (m: arbitrary integer of 1 to N) th voice processing device 20 uses the signal input by the mth microphone among the N microphones as the target voice signal, and excludes m from 1 to N. The signal input by the second microphone may be used as a reference signal.

As a result, the voice processing system 5 can suppress the crosstalk component for each of the sounds picked up by the plurality of microphones, and can improve the noise suppression accuracy of the crosstalk component and the like of the voice signal picked up by each microphone. ..

Further, the microphones MC1 to MC4 may be arranged in the vehicle interior.

As a result, the voice processing device 21 can suppress the cross talk component included in the voice spoken by the speaker even in a situation where a plurality of occupants are present in a narrow space, for example, in the vehicle interior.

Further, the voice processing system 5 may include a voice recognition engine 40 (an example of a voice recognition processing unit) that performs voice recognition processing on an output signal.

As a result, the voice processing system 5 improves the followability with respect to the update of the filter coefficient of the adaptive filter, so that the voice based on the output signal in the initial stage after the filter coefficient is changed (for example, after the speaker is changed in the vehicle interior). Voice recognition accuracy is improved. Therefore, the voice recognition accuracy is improved immediately after the speaker emits the voice, and the recognition accuracy of the operation information in the application using the voice recognition is improved. Therefore, the voice processing system 5 can smoothly give an instruction to an application that can be operated by voice recognition, for example.

(Second Embodiment)
The second embodiment shows a case where speaker detection is performed to detect a speaker who is speaking among the occupants in the vehicle interior, and the speaker detection result is used as an auxiliary for suppression processing of the crosstalk component.

In the second embodiment, the same components and operations as those described in the first embodiment will be omitted or simplified by using the same reference numerals.

FIG. 4 is a diagram showing a hardware configuration of the voice processing device 21A according to the second embodiment. Similar to the first embodiment, the voice processing system 5A of the second embodiment includes a plurality of (for example, four) microphones MC1 to MC4, a voice processing device 20A, and voice arranged in the vehicle interior of the vehicle 10. It has a configuration including a recognition engine 40 and a car navigation device 50. Further, the voice processing device 20A may be, for example, voice processing devices 21A, 22A, 23A, 24A. The voice processing devices 21A, 22A, 23A, and 24A all have the same configuration and function. Here, the voice processing device 21A will be mainly described.

The voice processing device 21A includes a voice input unit 29, a plurality of (for example, four) adaptive filters F2, F3, F4, F5, an adder 27, an adaptive filter control unit 28A, a storage unit 28B, and a signal detection unit. 30 and.

The signal detection unit 30 inputs the audio signal A from the microphone MC1 and the reference signals B, C, and D from the microphones MC2, MC3, and MC4, and based on the sound pressure level (signal level) of these signals, the signal detection unit 30 inputs. Detects the position of the speaker. For example, when the sound pressure level of the voice signal of the voice picked up by the microphones MC1 to MC4 is higher than the threshold value th2, it is determined that there is an occupant speaking into the microphone, and the position of the speaker is specified. Good. The signal detection unit 30 may notify the adaptive filter control unit 28A of the detection result of the speaker position. The signal detection unit 30 may be provided for each of the voice processing devices 21A to 24A, or may be provided for the entire voice processing system 5A.

The adaptive filter control unit 28A may update the adaptive filter corresponding to the speaker position detected by the signal detection unit 30 as an update target (control target), and update the filter coefficient of the adaptive filter. On the other hand, the adaptive filter control unit 28A does not have to update the adaptive filter corresponding to other than the detected speaker position, and does not have to update the filter coefficient of the adaptive filter.

For example, when an utterance is detected in the passenger seat, the adaptive filter control unit 28A may update the filter coefficient of the adaptive filter F3. Further, when the utterance is detected in both the left rear seat and the right rear seat, the adaptive filter control unit 28A may update the filter coefficient of the adaptive filter F2, assuming that there are a plurality of speakers.

When the accuracy of the speaker position detected by the signal detection unit 30 is low, the adaptive filter control unit 28A updates the adaptive filter F2 (including the three adaptive filters F2A, F2B, and F2C), and the pass signal of the adaptive filter F2. An output signal may be obtained based on E'. As a result, it is difficult to accurately detect the positions of a plurality of speakers, but even if the accuracy of the speaker positions is low and it is difficult to estimate the speaker positions, the voice processing device 21A suppresses a large deterioration in sound quality. it can.

On the other hand, when the accuracy of the speaker position detected by the signal detection unit 30 is high, the adaptive filter control unit 28A updates any of the adaptive filters corresponding to the speaker position and passes through the updated adaptive filter. An output signal may be obtained based on the passing signal. As a result, when the accuracy of the speaker position is high and the speaker position can be specified with high accuracy, the voice processing device 21A can sufficiently suppress the crosstalk component (for example, the speaker) contained in the voice signal A.

In this way, the voice processing device 21A can omit at least a part of the operations related to all the adaptive filters F2 to F5 by estimating the speaker position by detecting the speaker position. The operations related to the adaptive filters F2 to F5 may include operations related to the generation of each passing signal, operations related to the generation of each subtraction signal, and the like. In this way, by using the function of the signal detection unit 30 as an auxiliary, the processing load of the voice processing device 21A can be reduced.

Here, an example of deriving the speaker position and the accuracy of the speaker position will be described.

The signal detection unit 30 may derive the accuracy of the speaker position by various methods. For example, the signal detection unit 30 determines the accuracy of the speaker position according to whether or not the sound pressure level (signal level) of the voice picked up by the microphones MC1 to MC4 exceeds the threshold value th3 (> threshold value th2). You can do it. Further, the signal detection unit 30 may include a camera. In this case, the signal detection unit 30 may, for example, take an image of the vicinity of the mouth of the occupant and analyze the captured image to determine whether or not the occupant is speaking. When the speaker detected by the sound pressure level and the speaker analyzed based on the image captured by the camera match, the signal detection unit 30 may determine that the accuracy of speaker position detection is high.

Further, when a person riding in the vehicle 10 often sits in the same seat in the vehicle 10, the signal detection unit 30 may detect the speaker position using a voiceprint. In this case, the voiceprints of each occupant are registered in advance in the storage unit 28B, and when there is an utterance, the signal detection unit 30 refers to the voiceprints registered in the storage unit 28B and picks up the sound with the microphone corresponding to the seat. It may be determined whether or not it matches the voiceprint of the voice to be played. If they match, the signal detection unit 30 may determine that the seating position of the occupant corresponding to the voiceprint is highly likely to be the speaker position.

Further, the signal detection unit 30 determines the accuracy of speaker position detection according to the distance between the installation position of the microphones MC1 to MC4 and the driver hm1 or the occupant hm2 to hm4 in which the microphones MC1 to MC4 mainly collect sound. You may judge that it is expensive. The signal detection unit 30 may estimate the above distance based on the delay component of the audio signal picked up by the microphones MC1 to MC4. For example, when the delay component of the audio signal is large, it can be determined that the distance between the microphone and the speaker is long and the accuracy of speaker position detection is low. When the delay component of the audio signal is small, it can be determined that the distance between the microphone and the speaker is short and the accuracy of speaker position detection is high.

The adaptive filter control unit 28A may calculate the signal level of the subtraction signal (error signal) as a score, calculate the detection result of the speaker position as a score, and determine the output signal based on these scores. In this case, the adaptive filter control unit 28A may associate the speaker position with the subtraction signal corresponding to the passing signal of the adaptive filter corresponding to the speaker position detected as the detection result of the speaker position. For example, the adaptive filter control unit 28A may determine the subtraction signal having the highest (or lowest) sum of the scores of both as the output signal.

Therefore, in the voice processing system 5A, the adaptive filter can be updated more accurately by detecting the speaker position, and a high-quality voice signal in which the crosstalk component is further suppressed can be obtained.

In addition, the adaptive filter control unit 28A has a high reliability (accuracy) of the speaker position detection result when the period in which the reliability (accuracy) is high continues for the threshold value th4 or more, that is, when the utterance is long in a certain seat. The filter coefficient that has been continuously updated in a high state may be saved in the storage unit 28B as the filter coefficient of the adaptive filter corresponding to the seat. Further, the adaptive filter control unit 28A may update the saved filter coefficient at regular intervals. When it is determined that the speaker position estimated based on the signal detection unit 30 or the signal level of the error signal is the position of the seat corresponding to the adaptive filter in which the filter coefficient is saved in the storage unit 28B, the adaptive filter control is performed. The unit 28A may read the saved filter coefficient and set it as an adaptive filter. As a result, the voice processing device 21A effectively suppresses the voice component (crosstalk component) of the speaker in this seat immediately after the position corresponding to the adaptive filter in which the filter coefficient is saved becomes the speaker position. it can.

As described above, the voice processing device 21A is a speaker (driver hm1) that emits the voice of the target component (main component of the voice signal A) or the main component other than the target component (main component of the reference signals B, C, D). , The signal detection unit 30 for detecting the position of the occupant hm2 to hm4) may be provided. The adaptive filter control unit 28A may determine the adaptive filter to be controlled based on the detected position of the speaker.

As a result, the voice processing device 21A can improve the selection accuracy of the applied filter to be updated by using the result of the speaker position detection as an auxiliary.

Further, the voice processing device 21A may further include a storage unit 28B. The adaptive filter control unit 28A may derive (for example, calculate) the accuracy of the position of the speaker detected by the signal detection unit 30. When the accuracy is equal to or higher than the threshold value th4 (an example of the second threshold value), the storage unit 28B describes the position of the speaker detected by the signal detection unit 30 at the time t1 (an example of the first time) and this story. The filter coefficient of the applied filter corresponding to the position of the person may be associated and stored. When the signal detection unit 30 detects the same speaker position as the time t1 at the time t2 (an example of the second time) after the time t1, the adaptive filter control unit 28A corresponds to this speaker. The filter coefficient of the applied filter may be updated with the filter coefficient associated with the speaker's position stored in the storage unit 28B.

As a result, when the same speaker (or a combination of the same speakers) is emitting voice at a plurality of times t1 and t2, the voice processing device 21A uses the value of the filter coefficient that has been updated in the past and has a proven track record. Therefore, it can be expected that the suppression accuracy of the noise component of the voice signal can be improved as in the past. Further, although it is difficult for the voice processing device 21A to determine the speaker, the speaker position detection accuracy can be stably updated to a proven filter coefficient by limiting the accuracy to the threshold value th4 or more.

Although various embodiments have been described above with reference to the drawings, it goes without saying that the present disclosure is not limited to such examples. It is clear that a person skilled in the art can come up with various modifications or modifications within the scope of the claims, which naturally belong to the technical scope of the present disclosure. Understood.

In the above embodiment, the processor may be physically configured in any way. Further, if a programmable processor is used, the processing content can be changed by changing the program, so that the degree of freedom in processor design can be increased. The processor may be composed of one semiconductor chip, or may be physically composed of a plurality of semiconductor chips. When composed of a plurality of semiconductor chips, each control of the above embodiment may be realized by a separate semiconductor chip. In this case, it can be considered that one processor is composed of those plurality of semiconductor chips. Further, the processor may be composed of a member (capacitor or the like) having a function different from that of the semiconductor chip. Further, one semiconductor chip may be configured so as to realize the functions of the processor and other functions. Further, a plurality of processors may be configured by one processor.

In the above embodiment, it is mainly illustrated that the voice of the speaker is included as the voice signal and the reference signal, but the voice signal and the reference signal may broadly include various sounds. For example, the audio signal and the reference signal may broadly include music, environmental sounds, mechanical sounds, and other sounds.

In the above embodiment, each threshold value may be a fixed value or a variable value. Each threshold value may be a predetermined value or a value input via an operation unit included in the voice processing system.

The present disclosure is useful for a voice processing system, a voice processing device, a voice processing method, and the like that can improve the suppression accuracy of a noise component of a voice signal to be acquired even when the number and position of noise sources change.

5 Voice processing system 10 Vehicles 20, 21, 21A, 22, 23, 24 Voice processing device 27 Adder 28, 28A Adaptive filter control unit 28B Storage unit 29 Voice input unit 30 Signal detection unit 40 Voice recognition engine 50 Car navigation device F2 , F2A, F2B, F2C, F3, F4, F5 Adaptive filter hm1 Driver hm2, hm3, hm4 Crew MC1, MC2, MC3, MC4 Microphone

Claims (13)

A first voice acquisition unit that acquires a voice signal including a first voice component and a voice component other than the first voice component, and
A plurality of second audio acquisition units that acquire a plurality of reference signals including a second audio component and an audio component other than the second audio component, and
A first adaptive filter that passes two or more reference signals out of the plurality of reference signals to generate a first pass signal, and
A plurality of second adaptive filters that pass a single different reference signal among the plurality of reference signals to generate a plurality of second pass signals, and a plurality of second adaptive filters.
Based on the voice signal, the first passing signal, and the plurality of second passing signals, the adaptive filter to be controlled among the first adaptive filter and the plurality of second adaptive filters is determined. A control unit that controls the filter coefficient of the adaptive filter to be controlled,
A voice processing system equipped with. The control unit
The first passing signal is subtracted from the audio signal to generate the first subtracting signal.
A plurality of second subtraction signals are generated by subtracting different second passing signals from the audio signal.
The adaptive filter to be controlled is determined based on the signal levels of the first subtraction signal and the plurality of second subtraction signals.
The voice processing system according to claim 1. The control unit
Based on the signal levels of the first subtraction signal and the plurality of second subtraction signals, any one of the first subtraction signal and the plurality of second subtraction signals is determined as an output signal.
The adaptive filter of any one of the first adaptive filter and the plurality of second adaptive filters corresponding to the output signal is determined as the adaptive filter to be controlled.
The voice processing system according to claim 2. The control unit determines the adaptive filter corresponding to the subtraction signal having the minimum signal level among the first subtraction signal and the plurality of second subtraction signals as the adaptive filter to be controlled.
The voice processing system according to claim 2 or 3. The control unit sets an adaptive filter corresponding to the subtraction signal whose signal level is equal to or lower than the first threshold value among the first subtraction signal and the plurality of second subtraction signals as the adaptive filter to be controlled. decide,
The voice processing system according to claim 2 or 3. Further, a detection unit for detecting the position of the speaker emitting the first voice component or the second voice component is provided.
The control unit determines an adaptive filter to be controlled based on the position of the speaker.
The voice processing system according to any one of claims 1 to 5. With a storage unit,
The control unit derives the accuracy of the position of the speaker detected by the detection unit.
When the accuracy is equal to or higher than the second threshold value, the storage unit includes the position of the speaker detected by the detection unit at the first time and the filter coefficient of the applied filter corresponding to the position of the speaker. , Associate and remember,
When the detection unit detects the same speaker position as the first time at a second time after the first time, the control unit determines the applicable filter corresponding to the speaker. The filter coefficient is updated with the filter coefficient stored in the storage unit and associated with the speaker position detected by the detection unit.
The voice processing system according to claim 6. The second audio acquisition unit has directivity in the direction of the sound source that emits the second audio component in order to acquire the reference signal.
The voice processing system according to any one of claims 1 to 7. A plurality of the first adaptive filter, the plurality of second adaptive filters, and a voice processing device including the control unit are provided.
Each voice signal acquired by each control unit in a plurality of voice processing devices is different.
The combination of each reference signal acquired by each of the first adaptive filters and the second adaptive filters in the plurality of voice processing devices is different.
The voice processing system according to any one of claims 1 to 8. The first voice acquisition unit and the plurality of second voice acquisition units are arranged in the vehicle interior.
The voice processing system according to any one of claims 1 to 9. A voice recognition processing unit that performs voice recognition processing on the output signal is further provided.
The voice processing system according to claim 3. A control unit that acquires an audio signal including a first audio component and an audio component other than the first audio component, and
A plurality of adaptive filters that acquire a plurality of reference signals including a second audio component and an audio component other than the second audio component, and
With
The plurality of adaptive filters
A first adaptive filter that passes two or more reference signals out of the plurality of reference signals to generate a first pass signal, and
A plurality of second adaptive filters, which pass a different single reference signal among the plurality of reference signals and generate a plurality of second pass signals, are included.
The control unit adapts the controlled object among the first adaptive filter and the plurality of second adaptive filters based on the audio signal, the first pass signal, and the plurality of second pass signals. The filter is determined, and the filter coefficient of the adaptive filter to be controlled is controlled.
Voice processing device. Acquires an audio signal containing an audio component other than the first audio component and the first audio component,
Acquire a plurality of reference signals including a second audio component and an audio component other than the second audio component,
Two or more of the plurality of reference signals generate a first pass signal that has passed through the first adaptive filter.
A plurality of second pass signals that have passed through a plurality of second adaptive filters through which a different single reference signal among the plurality of reference signals passes are generated.
Based on the voice signal, the first passing signal, and the plurality of second passing signals, the adaptive filter to be controlled among the first adaptive filter and the plurality of second adaptive filters is determined. Controlling the filter coefficient of the adaptive filter to be controlled,
Voice processing method.

Images