Extension of the multi-channel Extension of the multi-channel
Wiener filter with localization cues Wiener filter with localization cues
for binaural noise reduction for binaural noise reduction
Simon Doclo
1,2, Rong Dong
2, Thomas J. Klasen
1,3, Jan Wouters
3, Simon Haykin
2, Marc Moonen
11Dept. of Electrical Engineering (ESAT-SCD), KU Leuven, Belgium
2Adaptive Systems Laboratory, McMaster University, Canada
3 Laboratory for Exp. ORL, KU Leuven, Belgium
WASPAA-2005, Oct 17 2005
Overview Overview
• Binaural hearing aids: noise reduction and preservation of binaural cues
• Overview of binaural noise reduction algorithms
• Binaural multi-channel Wiener filter:
o Estimate of speech component at both hearing aids
o Speech cues are preserved – noise cues may be distorted
• Preservation of binaural noise cues:
o Partial estimation of noise component o Extension with ITD and ILD cost function
• Experimental results
• Audio demonstration
Problem statement Problem statement
• Hearing impairment reduction of speech intelligibility in background noise
o Signal processing to selectively enhance useful speech signal o Many hearing impaired are fitted with hearing aid at both ears
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
PL PR
ITD
ILD
LR
P P
• Binaural auditory cues:
o Interaural Time Difference (ITD) – Interaural Level Difference (ILD) o Binaural cues, in addition to spectral and temporal cues, play an
important role in binaural noise reduction and sound localization
Problem statement Problem statement
• Bilateral system:
o Independent processing of left and right hearing aid
o Binaural cues may be distorted (cf. poster Tim Van den Bogaert)
• Binaural system:
o Cooperation between left and right hearing aid (e.g. wireless link) o Assumption: all microphone signals are available at the same time
Objectives/requirements for binaural algorithm:
1. SNR improvement: noise reduction, limit speech distortion
2. Preservation of binaural cues (speech/noise) to exploit binaural hearing advantage
3. No assumption about position of speech source and microphones
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Binaural noise reduction techniques Binaural noise reduction techniques
• Fixed beamforming: spatial selectivity + binaural speech cues
o Maximize directivity index while restricting ITD error o Superdirective beamformer using HRTFS
broadside array, limited performance, known geometry
[Desloge, 1997]
[Lotter, 2004]
• CASA-based techniques
o Computation and application of (real-valued) binaural mask based on binaural and temporal/spectral cues
mostly for 2 microphones, “spectral-subtraction”-like problems
[Kollmeier, Peissig, Wittkop, Dong, Haykin]
• Adaptive beamforming: based on GSC-structure
o Divide frequency spectrum: low-pass portion unaltered to preserve ITD cues, high-pass portion processed using GSC
low-pass: no noise reduction, high-pass: no cue preservation o TF-GSC: minimize output energy, constraint: output speech
component = speech component in reference microphone signal binaural noise cues may be distorted
[Welker, 1997]
[Gannot, 2001]
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Binaural noise reduction techniques Binaural noise reduction techniques
• Binaural multi-channel Wiener filter
o MMSE estimate of speech component in reference microphone signal at both ears
speech cues are preserved, no assumptions about position of speech source and microphones
noise cues may be distorted
[Doclo, Klasen, Wouters, Moonen]
Extension of MWF:
preservation of binaural speech and noise cues without comprimising noise reduction performance
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Binaural multi-channel Wiener filter Binaural multi-channel Wiener filter
• Configuration: microphone array at left and right hearing aid
0,m( ) = 0,m( ) 0,m( ), = 0 0 1 Y
X
V
m M 0,0 0, 0 1 1,0 1, 1 1
( ) =
Y ( )
Y M ( )
Y ( )
Y M ( )
TY
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
0( ) = 0H( ) ( ), 1( ) = 1H( ) ( ) , ( ) = 0T( ) 1T( ) T Z
W
Y
Z
W
Y
W
W
W
• Use all available microphone signals to compute output signal at
both ears computation of filters W
0and W
1Binaural multi-channel Wiener filter Binaural multi-channel Wiener filter
• SDW-MWF: estimate speech component in reference mic signal at both ears + trade-off noise reduction and speech distortion
2 2
,0 {| 0, 0 0H | } ( 0 1) {| 0H | }
SDW r
J E X W Y
E W Vr
0r
1,0 ,1
( ) = H H H
SDW SDW SDW
J W J J P W RW W r r W WSDW = R r1
0 0
0 1
1 1
= , =
x, =
x v M,
x y vx M x v
P P P
r R R 0
r R R R R
r 0 R R
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Preservation of binaural cues Preservation of binaural cues
• SDW-MWF: binaural speech cues are generally preserved, binaural noise cues may be distorted
• Partial estimation of noise component
o Estimate of sum of speech component and scaled noise component
considerable reduction of noise reduction performance
[Klasen, 2005]
2 ,0( 0) {| ( 0,0 0 0, 0) 0H | }
MSE r r
J W E X V W Y
• Extension of SDW-MWF with binaural cues
o Add term related to ITD and ILD cue to SDW cost function
o Link computation of filters W0 and W1
o Weight factors can be frequency-dependent
o Task: perceptually relevant expressions for binaural cues
( ) = ( )
tot SDW
J W J W
2
( )
( )
ITD
out des
J
ITD ITD
W
W
2 ( )
( )
ILD
out des
J
ILD ILD
W
W
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
ITD-ILD cost functions ITD-ILD cost functions
• ITD: phase of cross-correlation between two signals
o Output: input:
o Cost function: cosine of phase difference between cross-correlations
*
0 1 0 1
{ v v } = H v
E Z Z W R W 0, 1, 0 1
0 1
= { r *r } = v( , ) s E V V R r r
0 1 0 1
2 2 2 2
0 1 0 1
( ) ( )
( ) = 1 cos( ( )) = 1
( ) ( )
H H
R v R I v I
ITD H H
R I v R v I
s s
J
s s
W R W W R W
W W
W R W W R W
• ILD: power ratio of two signals
o Output: input:0 22 0 0
1 1 1
{| | }
= .
{| | }
H
v v
H
v v
E Z E Z
W R W W R W
2
0, 0 0 0
2
1,1 1 1
{| | } ( , )
{| | } = ( , )
r v
r v
E V r r
P E V R r r R
2
0 0
1 1
( ) =
H v
ILD H
v
J
P
W R W W W R W
• Other possibility: specifiy desired angle
vand use HRTFs
• Estimate noise cross-correlation/power during noise segments
2 2
*
0 1 0 1
( ) HRTF ( , ) HRTF ( , ),
v vHRTF ( , ) / HRTF ( , )
v vs
P
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Experimental results Experimental results
• Binaural recordings on KEMAR, 2 microphones on each hearing aid (d=1cm)
• Speech source in front, multi-talker babble noise at 45
• SNR=0 dB, f
s=16 kHz, FFT-size N=256,
0=
1=1
• Performance measures:
o SNR improvement (left/right)
o Mean ITD/ILD cost function (speech/noise component)
Noise 1
Reverberation time
= 125 msec
HA
RHA
LProblem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Experimental results Experimental results
• Partial estimation of noise component
o ITD and ILD cost function of speech and noise components decrease
o SNR improvement is significantly reduced
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0 2 4 6 8 10 12
SNR [dB]
Left ear Right ear
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
-4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0
J ITD [dB]
Noise component Speech component
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
Experimental results Experimental results
• Extension with ITD-ILD cost function
o ITD and ILD cost function of noise component decrease o SNR improvement is practically not reduced
o ITD and ILD cost function of speech component not degraded
1 0
2 x 10-3
0 0.5
1 -2.5
-2 -1.5 -1 -0.5 0
ITD cost function - noise component
J ITD [dB]
1 0
2 x 10-3
0 0.5
1 8.5
9 9.5 10 10.5
SNR improvement right ear
SNR [dB]
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions
• PS: Other demos available (=5, 2 noise sources, T =250 ms)
Audio demo Audio demo
Algorithm Sound files
SNRavg ITDnoITD
sp ILDnoILD
spInput
SDW-MWF
9.7 -0.43-1.86
-1.43-1.31
Partial estimation (=0.8)
1.8 -1.69-3.40
-3.33-2.73
ITD-ILD xcorr (=0.002,=1)
9.1 -2.47-1.80
-4.69-1.41
• Speech enhancement for binaural hearing aids:
o SNR improvement
o Preservation of binaural speech and noise cues
o No assumptions about position speech source and microphones
• Suitable noise reduction algorithms:
o Multi-channel Wiener filter (but also e.g. Transfer Function GSC) speech cues are preserved noise cues may be distorted
• Preservation of binaural noise cues:
o Extension with ITD and ILD cost functions
o Other possible extensions: Interaural Transfer Function (ITF)
• Future work:
o Multiple noise sources ?
o Better perceptual cost functions/performance measures o Tuning of frequency-dependent weight factors
Conclusions and future work Conclusions and future work
Problem statement
Binaural noise reduction
Multi-channel Wiener filter
Preservation of binaural cues
Experiments
Conclusions