Dereverberation and noise reduction method for microphone array and apparatus using the same

• US 8,351,618 B2
• Filed: 03/11/2010
• Issued: 01/08/2013
• Est. Priority Date: 12/25/2009
• Status: Active Grant

• Current Assignee: U-MEDIA Communications, Inc.
• Original Assignee: National Chiao Tung University (Government of The Republic of China)
• Inventors: Bai, Mingsian R., Hur, Ker-Nan, Liou, Ying-Ting
• Primary Examiner(s): Lee, Ping
• Application Number:
  US12/721,566
• Publication Number:
  US 20110158418A1
• Time in Patent Office: 1,034 Days
• Field of Search: 381/92, 381 941- 943, 381/94.7
• US Class Current: 381/92
• CPC Class Codes: H04B 3/23 using a replica of transmitted signal in the time domain, e.g. echo cancellers

Abstract

A dereverberation and noise reduction method adapted for a microphone array and an apparatus using the same are proposed. The microphone array receives a plurality of audio signals from an audio source. The dereverberation and noise reduction method includes the following steps. The received audio signals are processed by a beamforming processing, and a first audio signal is generated. Besides, the received audio signals are processed by a suppression processing, and a suppression audio vector is generated. Further, suppression audio vector is processed by an adaptive filtering processing, and a second audio signal is generated. In addition, the second audio signal is subtracted from the first audio signal to acquire an audio output signal, where parameters of the adaptive filtering processing are adjusted according to a feedback of the audio output signal.

Claims

1. A dereverberation and noise reduction method, adapted for a single input multiple output (SIMO) microphone array, wherein the SIMO microphone array comprises a plurality of microphones, and each of the microphones respectively receives one of a plurality of audio signals from an audio source, wherein each of the audio signals respectively experiences a gain of a system response between the audio source and the SIMO microphone array, the system response is obtained by a measurement in advance and expressed as a system response matrix H, and the gain respectively experienced by each of the audio signals is not substantially completely identical, the dereverberation and noise reduction method comprising:

applying a beamforming processing on the audio signals received by the microphone array, and generating a first audio signal, wherein the beamforming processing allows a main-lobe portion of the audio signals pass through and reduces a plurality of side-lobe portions of the audio signals, and the beamforming processing executes a computation of an inverse filter matrix C, wherein a relationship between the inverse filter matrix C and the system response matrix H is expressed as the following equation (1);

C^T=α

H^H 

 

equation (1),

wherein, α

is a real number and greater than zero, C^T is a transpose matrix of the inverse filter matrix C, and H^H is a Hermitian matrix of the system response matrix H;

applying a suppression processing on the audio signals received by the microphone array, and generating a suppression audio vector, wherein the suppression processing allows the side-lobe portions of the audio signals pass through and reduces the main-lobe portion of the audio signals, and the suppression processing executes a computation of a blocking matrix B, wherein the blocking matrix B is expressed as the following equation (2);

h^HB=0 

 

equation (2),

wherein, h^H is a Hermitian matrix of a frequency response matrix h between the audio source and each of the microphones;

applying an adaptive filtering processing on the suppression audio vector, and generating a second audio signal and

subtracting the second audio signal from the first audio signal to acquire an audio output signal, wherein a plurality of parameters of the adaptive filtering processing are adjusted according to a feedback of the audio output signal.

2. The dereverberation and noise reduction method according to claim 1, wherein after the beamforming processing, the dereverberation and noise reduction method further comprises:

applying an inverse Fast Fourier Transform (IFFT) processing on the audio signals processed by the beamforming processing; and

applying an adequate time-domain shifting processing on the audio signals processed by the beamforming processing and the IFFT processing so as to generate the first audio signal.

3. The dereverberation and noise reduction method according to claim 1, wherein the real number α

is 1.

4. The dereverberation and noise reduction method according to claim 1, wherein the real number α

is expressed as the following equation (3);

α

=(∥

H∥

₂²)<sup>−

1</sup> 

 

equation (3),

wherein, ∥

H∥

²₂ is a squared value of a 2-norm value of the system response matrix H.

5. The dereverberation and noise reduction method according to claim 1, wherein when the Hermitian matrix h^H of the system frequency response matrix h between the audio source and each of the microphones is expressed as the following equation (4), the blocking matrix B is expressed as the following equation (5):

6. The dereverberation and noise reduction method according to claim 1, wherein the adaptive filtering processing uses a least mean square method to process the suppression audio vector.

7. The dereverberation and noise reduction method according to claim 6, wherein the least mean square method is expressed as the following equation (6) to equation (9):

8. The dereverberation and noise reduction method according to claim 1, wherein each of the microphones comprises an omni-directional microphone.

9. A dereverberation and noise reduction method, adapted for a single input multiple output (SIMO) microphone array, wherein the SIMO microphone array comprises a plurality of microphones, and each of the microphones respectively receives one of a plurality of audio signals from an audio source, wherein each of the audio signals respectively experiences a gain of a system response between the audio source and the SIMO microphone array, the system response is obtained by a measurement in advance and expressed as a system response matrix H, and the gain respectively experienced by each of the audio signals is not substantially completely identical, the dereverberation and noise reduction method comprising:

applying a first signal processing on the audio signals received by the microphone array, and generating a first audio signal, wherein the first signal processing allows a main-lobe portion of the audio signals pass through and reduces a plurality of side-lobe portions of the audio signals, and the first signal processing executes a computation of an inverse filter matrix C, wherein a relationship between the inverse filter matrix C and the system response matrix H is expressed as the following equation (1);

C^T=α

H^H 

 

equation (1),

wherein, α

is a real number and greater than zero, C^T is a transpose matrix of the inverse filter matrix C, and H^H is a Hermitian matrix of the system response matrix H;

applying a second signal processing on the audio signals received by the microphone array, and generating a suppression audio vector, wherein the second signal processing allows the side-lobe portions of the audio signals pass through and reduces the main-lobe portion of the audio signals;

applying a third signal processing on the suppression audio vector, and generating a second audio signal; and

subtracting the second audio signal from the first audio signal to acquire an audio output signal, wherein a plurality of parameters of the third signal processing are adjusted according to a feedback of the audio output signal.

10. The dereverberation and noise reduction method according to claim 9, the second signal processing executes a computation of a blocking matrix B, wherein the blocking matrix B is expressed as the following equation (2):

h^HB=0 

 

equation (2),

wherein, h^H is a Hermitian matrix of a frequency response matrix h between the audio source and each of the microphones.

11. The dereverberation and noise reduction method according to claim 10, wherein after the first signal processing, the dereverberation and noise reduction method further comprises:

applying an inverse Fast Fourier Transform (IFFT) processing on the audio signals processed by the first signal processing; and

applying an adequate time-domain shifting processing on the audio signals processed by the first signal processing and the IFFT processing so as to generate the first audio signal.

12. The dereverberation and noise reduction method according to claim 10, wherein when the Hermitian matrix h^H of the system frequency response matrix h between the audio source and each of the microphones is expressed as the following equation (4), the blocking matrix B is expressed as the following equation (5):

13. The dereverberation and noise reduction method according to claim 9, wherein the real number α

is 1.

14. The dereverberation and noise reduction method according to claim 9, wherein the real number α

is expressed as the following equation (3);

α

=(∥

H∥

₂²)<sup>−

1</sup> 

 

equation (3),

wherein, ∥

H∥

₂² is a squared value of a 2-norm value of the system response matrix H.

15. The dereverberation and noise reduction method according to claim 9, wherein the third signal processing uses a least mean square method to process the suppression audio vector.

16. The dereverberation and noise reduction method according to claim 15, wherein the least mean square method is expressed as the following equation (6) to equation (9):

17. The dereverberation and noise reduction method according to claim 9, wherein each of the microphones comprises an omni-directional microphone.

18. A dereverberation and noise reduction method, adapted for a single input multiple output (SIMO) microphone array, wherein the SIMO microphone array comprises a plurality of microphones, and each of the microphones respectively receives one of a plurality of audio signals from an audio source, wherein each of the audio signals respectively experiences a gain of a system response between the audio source and the SIMO microphone array, the system response is obtained by a measurement in advance and expressed as a system response matrix H, and the gain respectively experienced by each of the audio signals is not substantially completely identical, the dereverberation and noise reduction method comprising:

applying a first signal processing on the audio signals received by the microphone array, and generating a first audio signal, wherein the first signal processing allows a main-lobe portion of the audio signals pass through and reduces a plurality of side-lobe portions of the audio signals;

applying a second signal processing on the audio signals received by the microphone array, and generating a suppression audio vector, wherein the second signal processing allows the side-lobe portions of the audio signals pass through and reduces the main-lobe portion of the audio signals, and the second signal processing executes a computation of a blocking matrix B, wherein the blocking matrix B is expressed as the following equation (1);

h^HB=0 

 

equation (1),

wherein, h^H is a Hermitian matrix of a frequency response matrix h between the audio source and each of the microphones;

applying a third signal processing on the suppression audio vector, and generating a second audio signal; and

subtracting the second audio signal from the first audio signal to acquire an audio output signal, wherein a plurality of parameters of the third signal processing are adjusted according to a feedback of the audio output signal.

19. The dereverberation and noise reduction method according to claim 18, wherein the first signal processing executes a computation of an inverse filter matrix C, and a relationship between the inverse filter matrix C and the system response matrix H is expressed as the following equation (2):

C^T=α

H^H 

 

equation (2),

wherein, α

is a real number and greater than zero, C^T is a transpose matrix of the inverse filter matrix C, and H^H is a Hermitian matrix of the system response matrix H.

20. The dereverberation and noise reduction method according to claim 19, wherein the real number α

is 1.

21. The dereverberation and noise reduction method according to claim 19, wherein the real number α

is expressed as the following equation (3);

α

=(∥

H∥

₂²)<sup>−

1</sup> 

 

equation (3),

wherein, ∥

H∥

₂² is a squared value of a 2-norm value of the system response matrix H.

22. The dereverberation and noise reduction method according to claim 18, wherein after the first signal processing, the dereverberation and noise reduction method further comprises:

applying an inverse Fast Fourier Transform (IFFT) processing on the audio signals processed by the first signal processing; and

applying an adequate time-domain shifting processing on the audio signals processed by the first signal processing and the IFFT processing so as to generate the first audio signal.

23. The dereverberation and noise reduction method according to claim 18, wherein when the Hermitian matrix h^H of the system frequency response matrix h between the audio source and each of the microphones is expressed as the following equation (4), the blocking matrix B is expressed as the following equation (5):

h^H=[a₁, a₂, . . . , a_n] 

 

equation (4),

wherein, a₁, a₂, . . . , a_n respectively is a frequency response value between the audio source and each of the microphones, and a₁, a₂, . . . a_n are real numbers, and n is a total number of the microphones.

24. The dereverberation and noise reduction method according to claim 18, wherein the third signal processing uses a least mean square method to process the suppression audio vector.

25. The dereverberation and noise reduction method according to claim 24, wherein the least mean square method is expressed as the following equations (6) to equation (9):

W_n(k+1)=W_n(k)+μ

×

y(k)Z_n(k) 

 

equation (7),

wherein,

W_n(k)=[w_n,0(k), w_n,1(k), . . . , w<sub>n,M2−

1</sub>(k)]^T 

 

equation (8),

Z_n(k)=[z_n(k), z_n(k−

1), . . . , z_n(k−

M₂+1)]^T 

 

equation (9),

wherein y(k) is the audio output signal when time is k, b(k) is the first audio signal when the time is k, L₁ is a time delay between the audio output signal and the first audio signal, W(k) is an adaptive filter matrix used by the third signal processing when the time is k, Z(k) is the suppression audio vector when the time is k, M₂ is a duration of the third signal processing and influences an overall operation efficiency of the dereverberation and noise reduction method, and μ

is a step size.

26. The dereverberation and noise reduction method according to claim 18, wherein each of the microphones comprises an omni-directional microphone.

27. A dereverberation and noise reduction apparatus, adapted for a single input multiple output (SIMO) microphone array, wherein the SIMO microphone array comprises a plurality of microphones, and each of the microphones respectively receives one of a plurality of audio signals from an audio source, wherein each of the audio signals respectively experiences a gain of a system response between the audio source and the SIMO microphone array, the system response is obtained by a measurement in advance and expressed as a system response matrix H, and the gain respectively experienced by each of the audio signals is not substantially completely identical, the dereverberation and noise reduction method comprising:

a fixed beamformer unit, configured for applying a beamforming processing on the audio signals received by the microphone array, and generating a first audio signal, wherein the beamforming processing allows a main-lobe portion of the audio signals pass through and reduces a plurality of side-lobe portions of the audio signals, and the beamforming processing executes a computation of an inverse filter matrix C, wherein a relationship between the inverse filter matrix C and the system response matrix H is expressed as the following equation (1);

C^T=α

H^H 

 

equation (1),

wherein, α

is a real number and greater than zero, C^T is a transpose matrix of the inverse filter matrix C, and H^H is a Hermitian matrix of the system response matrix H;

a blocking matrix processing unit, configured for applying a suppression processing on the audio signals received by the microphone array, and generating a suppression audio vector, wherein the suppression processing allows the side-lobe portions of the audio signals pass through and reduces the main-lobe portion of the audio signals, and the suppression processing executes a computation of a blocking matrix B, wherein the blocking matrix B is expressed as the following equation (2);

h^HB=0 

 

equation (2),

wherein, h^H is a Hermitian matrix of a frequency response matrix h between the audio source and each of the microphones;

a multiple input canceller unit, coupled to the blocking matrix processing unit and configured for applying an adaptive filtering processing to the suppression audio vector, and generating a second audio signal; and

a computation unit, coupled to the fixed beamformer unit and the multiple input canceller unit, wherein the computation unit comprises a first computation input terminal, a second computation input terminal and a computation output terminal, wherein the first computation input terminal receives the first audio signal, the second computation input terminal receives the second audio signal, and the computation unit is configured for subtracting the second audio signal from the first audio signal to acquire an audio output signal, wherein a plurality of parameters of the multiple input canceller unit is adjusted according to a feedback of the audio output signal.

28. The dereverberation and noise reduction apparatus according to claim 27, wherein after the beamforming processing, the dereverberation and noise reduction apparatus further comprises:

an Inverse Fast Fourier Transform (IFFT) processing unit, configured for applying an IFFT processing on the audio signals processed by the beamforming processing; and

the fixed beamforming unit, configured for applying an adequate time-domain shifting processing on the audio signals processed by the beamforming processing and the IFFT processing so as to generate the first audio signal.

29. The dereverberation and noise reduction apparatus according to claim 27, wherein the real number α

is 1.

30. The dereverberation and noise reduction apparatus according to claim 27, wherein the real number α

is expressed as the following equation (3);

α

=(∥

H∥

₂²)<sup>−

1</sup> 

 

equation (3),

wherein, ∥

H∥

₂² is a squared value of a 2-norm value of the system response matrix H.

31. The dereverberation and noise reduction apparatus according to claim 27, wherein when the Hermitian matrix h^H of the system frequency response matrix h between the audio source and each of the microphones is expressed as the following equation (4), the blocking matrix B is expressed as the following equation (5):

32. The dereverberation and noise reduction apparatus according to claim 27, wherein the adaptive filtering processing uses a least mean square method to process the suppression audio vector.

33. The dereverberation and noise reduction apparatus according to claim 32, wherein the least mean square method is expressed as the following equations (6) to equation (9):

34. The dereverberation and noise reduction apparatus according to claim 27, wherein each of the microphones comprises an omni-directional microphone.