Method and device for suppressing residual echoes based on inverse transmitter receiver distance and delay for speech signals directly incident on a transmitter array

Patent metadata

Field	Value
Publication number	US-9685172-B2
Application number	US-201113642661-A
Country	US
Kind code	B2
Filing date	Oct 24, 2011
Priority date	Jul 8, 2011
Publication date	Jun 20, 2017
Grant date	Jun 20, 2017

Abstract

The present invention discloses a method and a device for suppressing residual echoes. The method comprises: performing adaptive filtering on M transmitter signals respectively to obtain M adaptive filtered signals; performing array-filtering 5 on the M−1 adaptive filtered signals other than the first adaptive filtered signal to obtain M−1 array-filter output signals by considering relative positions of the receiver and each of the transmitters and the time delay attributed to distances between the transmitters and the receiver; subtracting each of the M−1 array-filter output signals from the first adaptive filtered signal respectively to obtain M−1 difference signals, performing time-domain/frequency-domain conversion on the M−1 difference signals respectively and selecting one of the frequency-domain signals that has the least energy; performing time-domain/frequency-domain conversion on the first adaptive filtered signal and the M th adaptive filtered signal and then performing speech probability filtering on the converted first adaptive filtered signal and the converted M th adaptive filtered signal to obtain one frequency-domain speech probability signal; and multiplying the frequency-domain speech probability signal with the selected signal that has the least energy, and performing frequency-domain/time-domain conversion on the multiplication result to obtain a signal as a transmitter output signal. The technical solutions of the present invention can suppress the residual echoes effectively without impairing near end speech.

First claim

The invention claimed is:

1. A method for suppressing residual echoes, the method being suitable for use in a communication apparatus comprising M transmitters and one receiver, wherein M is a natural number greater than 1, and the M transmitters are arranged in line to form an array, the method comprising:
performing adaptive filtering on M transmitter signals respectively relative to a receiver signal to obtain M adaptive filtered signals;
processing M−1 adaptive filtered signals except the first adaptive filtered signal by respective array-filters to obtain M−1 array-filter output signals, wherein for each of the M−1 adaptive filtered signals, weight of the array-filter that corresponds to the adaptive filtered signal is determined according to a relative positions between the receiver and first transmitter and a relative position between the receiver and one of transmitters that corresponds to current adaptive filtered signal;
subtracting each of the M−1 array-filter output signals from the first adaptive filtered signal respectively to obtain M−1 difference signals, performing time-domain/frequency-domain conversion on the M−1 difference signals respectively and selecting a frequency-domain signal that has the least energy;
performing time-domain/frequency-domain conversion on the first adaptive filtered signal and the M th adaptive filtered signal and then performing speech probability filtering on a converted first adaptive filtered signal and a converted M th adaptive filtered signal to obtain one frequency-domain speech probability signal; and
multiplying the frequency-domain speech probability signal with the selected frequency-domain signal that has the least energy, and performing frequency-domain/time-domain conversion on the multiplication result to obtain a signal as a speech output signal in which the residual echoes have been suppressed;
wherein the step of, for each of the M−1 adaptive filtered signals, weight of the array-filter that corresponds to the adaptive filtered signal is determined according to a relative positions between the receiver and first transmitter and a relative position between the receiver and one of the transmitters that corresponds to current adaptive filtered signal comprises:
for each of current adaptive filtered signals among the M−1 adaptive filtered signals except the first adaptive filtered signal, weights of the respective array-filter are determined according to the following formulas:

∂

E
⁡

[

(

e
1

-

e
2

*
h

)

2

]

∂
h

=
0

,

⁢

e
1

=

D
2

D
1

⁢

e
2

⁡

(

t
+

D
2

c

+

D
1

c

)

where E[.] represents an averaging operation, h represents the array-filter, e 1 represents the first adaptive filtered signal, e 2 represents a current adaptive filtered signal, D 1 represents a distance between the receiver and the first transmitter, D 2 represents a distance between the receiver and one of the transmitters that corresponds to the current adaptive filtered signal, and c represents the acoustic speed, t represents a current time;
wherein the step of performing time-domain/frequency-domain conversion on the first adaptive filtered signal and the M th adaptive filtered signal and then performing speech probability filtering on a converted first adaptive filtered signal and a converted M th adaptive filtered signal to obtain one frequency-domain speech probability signal comprises:
calculating an arriving angle α(f) of a sound signal arriving at the transmitter array at each frequency point according to the two results of performing time-domain/frequency-domain conversion on the first adaptive filtered signal and the M th adaptive filtered signal; and
then calculating the speech probability pF(f) according to the following formula:
pF ( f )=1−α( f )/90;

where pF(f) is the frequency-domain speech probability signal obtained, f represents the frequency.

2. The method of claim 1 , further comprising a following step after the step of multiplying the frequency-domain speech probability signal with the selected frequency-domain signal that has the least energy:
performing spectrum filtering on the multiplication result and a receiver signal that has been subject to time-domain/frequency-domain conversion to obtain a spectrum filtering result, subtracting the spectrum filtering result from the multiplication result to obtain a subtraction result, and performing frequency-domain/time-domain conversion on the subtraction result to obtain a signal as the speech output signal in which the residual echoes have been suppressed.

3. The method of claim 2 , wherein the step of performing spectrum filtering on the multiplication result and a receiver signal that has been subject to time-domain/frequency-domain conversion to obtain a spectrum filtering result comprises:
dividing the full frequency range into N subbands with boundaries B 1 ˜BN+1, and performing the following calculations in each of the N subbands:
calculating a matching function HM(f):

H
M

⁡

(
f
)

=

∑

f

Assignees

• Lou Shasha
• Li Bo
• Liu Song
• Goertek Inc

Inventors

• Lou Shasha
• Li Bo
• Liu Song

Classifications

• G10L21/0216 Primary

  characterised by the method used for estimating noise · CPC title

• H04R2499/13

  Acoustic transducers and sound field adaptation in vehicles · CPC title

• G10L15/20

  Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) · CPC title

• G10L2021/02082

  the noise being echo, reverberation of the speech · CPC title

• H04R3/005

  for combining the signals of two or more microphones (specially adapted for hearing aids H04R25/407) · CPC title