Design and low-cost Design and low-cost
implementation of a robust implementation of a robust multichannel noise reduction multichannel noise reduction scheme for cochlear implants scheme for cochlear implants
Simon Doclo
1, Ann Spriet
1,2, Jean-Baptiste Maj
1,2,
Marc Moonen
1, Jan Wouters
2, Bas Van Dijk
3, Jan Janssen
31
Dept. of Electrical Engineering (ESAT-SCD), KU Leuven, Belgium
2
Laboratory for Exp. ORL, KU Leuven, Belgium
3
Cochlear Technology Centre Europe, Belgium
DARTS, 22 October 2003
2
Overview Overview
• Problem statement : hearing in background noise
• Adaptive beamforming: GSC
o not robust against model errors
• Design of robust noise reduction algorithm
o robust fixed spatial preprocessor o robust adaptive stage
• Experimental results
• Low-cost implementation of adaptive stage
o stochastic gradient algorithms
o computational complexity + memory requirements
• Conclusions
33
Problem statement Problem statement
• Hearing problems effect more than 12% of population
• Digital hearing instruments allow for advanced signal processing, resulting in improved speech understanding
• Major problem: (directional) hearing in background noise
o reduction of noise wrt useful speech signal o multiple microphones + DSP in BTE
o current systems: simple fixed and adaptive beamforming o robustness important due to small inter-microphone distance
hearing aids and cochlear implants
design of robust multi-microphone noise reduction scheme
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
4
Cochlear implants Cochlear implants
• Working principle: sound is converted to electrical stimuli in speech processor, allowing deaf people to hear again
cochlea
brain
middle ear sound
external ear
multi-channel electrode stimulator
box speech
processor
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
55
Algorithmic requirements Algorithmic requirements
• ‘Blind’ techniques: unknown noise sources and acoustic environment
• Adaptive: time-variant signals and acoustic environment
• Robustness:
o microphone characteristics (gain, phase, position)
o other deviations from assumed signal model (e.g. VAD)
• Implementation issues:
o number of microphones
o low computational complexity o memory
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
6
State-of-the art noise reduction State-of-the art noise reduction
• Single-microphone techniques:
o spectral subtraction, Kalman filter, subspace-based
o only temporal and spectral information limited performance
• Multi-microphone techniques:
o exploit spatial information
o Fixed beamforming: fixed directivity pattern
o Adaptive beamforming (e.g. GSC) : adapt to different acoustic environments improved performance
o Multi-channel Wiener filtering (MWF): MMSE estimate of speech component in microphones improved robustness
Sensitive to a-priori assumptions
Robust scheme, encompassing both GSC and MWF
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
77
Adaptive beamforming: GSC Adaptive beamforming: GSC
• Fixed spatial preprocessor:
o Fixed beamformer creates speech reference o Blocking matrix creates noise references
• Adaptive noise canceller:
o Standard GSC minimises output noise power
Spatial preprocessing
]
0[k u
]
1[k u
]
1[k uN Fixed
beamformer A(z) Speech
reference ]
0[k y
Blocking matrix
B(z) Noise
references ]
1[k y
]
2[k y
]
1[k yN
Adaptive Noise Canceller
] [k z
]
1[k w
]
2[k w
]
1[k wN (adaptation during noise)
]
0
[ k y
noise speech
] [ ]
[ ]
[ k x k v k y
i
i
i
0 2
]
[
[ ] [ ] [ ]
min E v k
Tk k
k
w v
w
TN
TT T
T T N T
T
k k
k k
k k
k k
] [ ]
[ ]
[ ]
[
] [ ]
[ ]
[ ]
[
1 2
1
1 2
1
v v
v v
w w
w w
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
8
Robustness against model errors Robustness against model errors
• Spatial preprocessor and adaptive stage rely on assumptions (e.g. no microphone mismatch, no reverberation,…)
• In practice, these assumptions are often not satisfied
o Distortion of speech component in speech reference o Leakage of speech into noise references, i.e.
• Design of robust noise reduction algorithm:
1. Design of robust spatial preprocessor (fixed beamformer) using statistical knowledge about microphone characteristics 2. Design of robust adaptive stage by taking speech distortion
into account in optimisation criterion speech distortion weighted multichannel Wiener filter (SDW MWF)
Speech component in output signal gets distorted ]
0
[ k x 0 x [k ]
] [ ] [ ]
[ ]
[ k x
0k k k
z
x w
Tx
Limit distortion both in and x
0[ k ] w
T[ k ] x [ k ]
Introduction
-Problem statement -State-of-the-art -GSC
Fixed beamforming
Adaptive stage
Implementation
Conclusions
99
Design of fixed beamformer Design of fixed beamformer
• FIR filter-and-sum structure: arbitrary spatial directivity pattern for arbitrary microphone configuration
• Objective: calculate fixed FIR filters w
n[k] such that
beamformer performs desired spatial and spectral filtering
Far-field:
- planar waves - equal attenuation
Spatial directivity pattern: ( , )
) (
) , ) (
,
(
w
Tg
S
H Z
Desired spatial directivity pattern: D ( , )
Introduction
Fixed beamforming -Broadband design -Robustness
Adaptive stage
Implementation
Conclusions
10
Design procedures Design procedures
• Design filter w such that spatial directivity pattern optimally fits minimisation of cost function
o Broadband problem: no design for separate frequencies
i design over complete frequency-angle region
• Cost functions:
o Least-squares quadratic function
o Non-linear cost function iterative optimisation = complex!
o Eigenfilter based on TLS-criterion GEVD
F H D d d J
LS(w ) ( , ) ( , ) ( , )
2amplitude and phase
) , ( H
) , ( D
F H D d d J
NL(w ) ( , ) ( , )
2( , )
2 2only amplitude
H D d d
F
J
tote TLS T
1 ) , ( )
, ) (
, ( )
(
2
w Q w w
Introduction
Fixed beamforming -Broadband design -Robustness
Adaptive stage
Implementation
Conclusions
1111
Non-linear procedure TLS-Eigenfilter
Simulations Simulations
Angle (deg) Freq (Hz)
dB
Angle (deg) Freq (Hz)
dB
Parameters:
-N=5, d=4cm -L=20, fs=8kHz -Pass: 40o-80o -Stop: 0o-30o + 90o-180o
Delay-and-sum
Angle (deg) Freq (Hz)
dB
12
• Small deviations from assumed microphone characteristics (gain, phase, position) large deviations from desired directivity
pattern, especially for small-size microphone arrays
• In practice microphone characteristics are never exactly known
• Consider all feasible microphone characteristics and optimise o average performance using probability as weight
– requires statistical knowledge about probability density functions
o worst-case performance minimax optimisation problem
– finite grid of microphone characteristics high complexity
Robust broadband beamforming Robust broadband beamforming
1 0
1 0
1
0
, , ) ( ) ( )
(
0 1
N N NA A
mean
J A A f A f A dA dA
J
N
Incorporate specific (random) deviations in design
position/ cos phase
) , ( gain
) , ( )
,
(
n j j f cn
a e
ne
n sA
Measurement or calibration procedure
Introduction
Fixed beamforming -Broadband design -Robustness
Adaptive stage
Implementation
Conclusions
1313
Simulations Simulations
• Non-linear design procedure
• N=3, positions: [-0.01 0 0.015] m, L=20, f
s=8 kHz
• Passband = 0
o-60
o, 300-4000 Hz (endfire) Stopband = 80
o-180
o, 300-4000 Hz
• Robust design - average performance:
Uniform pdf = gain (0.85-1.15) and phase (-5
o-10
o)
• Deviation = [0.9 1.1 1.05] and [5
o-2
o5
o]
Design J J
devJ
meanJ
maxNon-robust 0.1585 87.131 275.40 3623.6
Average cost 0.2196 0.2219 0.3371 0.4990 Maximum
cost 0.1707 0.1990 0.4114 0.4167
Introduction
Fixed beamforming -Broadband design -Robustness
Adaptive stage
Implementation
Conclusions
14
Non-robust design Robust design
No deviationsDeviations (gain/phase)
Simulations Simulations
Angle
(deg) Frequency
(Hz)
dB
Angle
(deg) Frequency
(Hz)
dB
Angle
(deg) Frequency
(Hz)
dB
Angle
(deg) Frequency
(Hz)
dB
Introduction
Fixed beamforming -Broadband design -Robustness
Adaptive stage
Implementation
Conclusions
Non-robust design Robust design
Simulations Simulations
1515
16
Design of robust adaptive stage Design of robust adaptive stage
• Distorted speech in output signal:
• Robustness: limit by controlling adaptive filter
o Quadratic inequality constraint (QIC):
= conservative approach, constraint f (amount of leakage) o Take speech distortion into account in optimisation criterion
– 1/ trades off noise reduction and speech distortion
• 1/ = 0 or no speech leakage GSC
• 1/ = 1 MMSE estimate of speech component in speech reference signal
– Regularisation term ~ amount of speech leakage
] [ ] [ ]
[ ]
[ k x
0k k k
z
x w
Tx ]
[ ] [ k k
T
x
w w [k ]
]
[k w
0 2 2
]
[
1 [ ] [ ]
] [ ] [ ]
[
min E v k
Tk k E
Tk k
k
w v w x
w
noise reduction speech distortion
Limit speech distortion, while not affecting noise reduction performance in case of no model errors QIC
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
1717
Wiener solution Wiener solution
• Optimisation criterion:
• Problem: clean speech and hence speech correlation matrix are unknown!
Approximation:
• VAD (voice activity detection) mechanism required!
0 2 2
]
[
1 [ ] [ ]
] [ ] [ ]
[
min E v k
Tk k E
Tk k
k
w v w x
w
[ ] [ ] [ ] [ ] [ ] [ ]
] 1
[
01
k v k E k
k E k
k E
k x x
Tv v
Tv
w
] [k
[ k ] x [ k ]
E x x
T [ k ] [ k ] E [ k ] [ k ] E [ k ] [ k ]
E x x
T y y
T v v
T1
11 1 ]
[
k
w E x [ k ] x
T[ k ] E v [ k ] v
T[ k ] E v [ k ] v
0[ k ]
speech-and-noise periods noise-only periods
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
18
Spatially-preprocessed SDW-MWF (1) Spatially-preprocessed SDW-MWF (1)
• In new optimisation criterion additional filter on speech reference signal may be added
]
0[k w
Spatial preprocessing
]
0[k u
]
1[k u
]
1[k uN Fixed
beamformer A(z) Speech
reference ]
0[k y
Blocking matrix
B(z) Noise
references ]
1[k y
]
2[k y
]
1[k yN Multi-channel Wiener Filter
(SDW-MWF)
] [k z
]
1[k w
]
2[k w
]
1[k wN
]
0
[ k w
0 2 2
]
[
1 [ ] [ ]
] [ ] [ ]
[
min E v k
Tk k E
Tk k
k
w v w x
w
TN
TT
T
k k k
k ] [ ] [ ] [ ]
[ w
0w
1w
1w
Speech Distortion Weighted Multichannel Wiener Filter (SDW-MWF)
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
1919
Spatially-preprocessed SDW-MWF (2) Spatially-preprocessed SDW-MWF (2)
• SP-SDW-MWF encompasses both GSC and SDW-MWF as special cases:
o No filter on speech reference
speech distortion regularised GSC (SDR-GSC)
– regularisation term added to GSC: the larger the speech leakage, the larger the regularisation
– special case: 1/ = 0 corresponds to traditional GSC
– SDR-GSC outperforms GSC with quadratic inequality constraint o Filter on speech reference
SDW-MWF on pre-processed microphone signals
– in absence of model errors = cascade of GSC + single-channel postfilter (SDW Wiener filter)
– Model errors do not effect its performance!
]
0
[ k w
]
0
[ k w
Outperforms QIC-GSC and SDR-GSC
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
20
Experimental validation (1) Experimental validation (1)
• Set-up:
o 3-mic BTE mounted on dummy head in office room (d = 1cm, 1.5cm) o Speech source in front of dummy head (90)
o 5 stationary speech-like noise sources: 75, 120, 180, 240, 285
o Microphone gain mismatch at 2
ndmicrophone
• Performance measures:
o Intelligibility-weighted signal-to-noise ratio
– Ii = band importance of i th one-third octave band
– SNRi = signal-to-noise ratio in i th one-third octave band o Intelligibility-weighted spectral distortion
– SDi = average spectral distortion in i th one-third octave band
2
i I
i
I SNR
iSNR
1 intellig
i I
i
I SD
iSD
1 intellig
c if
f x
i
f
df f
i
G
c i c
, 6 / 1 6
/ 1 2
2 10
2 2
) ( log
10 SD
6 , / 1
6 , / 1
G
x( f ) E E X Z
x22( ( f f ) )
(Power Transfer Function for speech component)
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
2121
Experimental validation (2) Experimental validation (2)
• SDR-GSC:
o GSC (1/ = 0) : degraded performance if significant leakage
o 1/ > 0 increases robustness (speech distortion noise reduction)
• SP-SDW-MWF:
o No mismatch: same as SDR-GSC, larger due to SDW-WF post-filter
o Performance is not degraded by mismatch
0 w
0
0 w
0
intellig
SNR SD
intelligIntroduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
22
Experimental validation (3) Experimental validation (3)
• GSC with QIC ( ) : QIC increases robustness GSC
• For large mismatch: less noise reduction than SP-SDW-MWF
]
[k w
QIC f (amount of speech leakage) less noise reduction than SDR-GSC for small mismatch
SP-SDW-MWF achieves better noise reduction than QIC-GSC, for a given maximum speech distortion
level
Introduction
Fixed beamforming
Adaptive stage -SP SDW MWF -Experimental
validation
Implementation
Conclusions
2323
Low-cost implementation (1) Low-cost implementation (1)
• Algorithms (in decreasing order of complexity):
o GSVD-based – chic et très cher
o QRD-based, fast QRD-based – chic et moins cher o Stochastic gradient algorithms – chic et pas cher
• Stochastic gradient algorithm (time-domain) :
o Cost function
results in LMS-based updating formula
o Allows transition to classical LMS-based GSC by tuning some parameters (1/, w
0)
0 2 1 [ ] [ ] 2
] [ ] [ ]
[ )
( E v k k k E k k
J w w
Tv w
Tx
[ ] [ ] [ ] [ ]
] [ ]
1
[ k w k v k v
0k v
Tk w k
w
regularisation term
] [ ] [ ] 1 [
k k
k x
Tw
x
Classical GSC
Introduction
Fixed beamforming
Adaptive stage
Implementation -Stochastic gradient -Complexity and
memory
Conclusions
24
Low-cost implementation (2) Low-cost implementation (2)
• Stochastic gradient algorithm (time-domain):
o Regularisation term is unknown
– Store samples in memory buffer during speech-and-noise periods and approximate regularisation term
– Better estimate of regularisation term can be obtained by smoothing (low-pass filtering)
• Stochastic gradient algorithm (frequency-domain):
o Block-based implementation: improve gradient estimate by averaging over K samples
o Frequency-domain: fast convolution and fast correlation Complexity reduction
Tuning of and 1/ per frequency
Still large memory requirement due to data buffers
o Approximations allow to replace data buffers by correlation matrices in frequency-domain memory reduction
] [ ] [ ] 1 [
k k
k x
Tw
x
Large buffer required
Introduction
Fixed beamforming
Adaptive stage
Implementation -Stochastic gradient -Complexity and
memory
Conclusions
2525
Complexity + memory Complexity + memory
• Parameters:
N = M = 2 (mics, adaptive filters), L = 64, fs= 16kHz, L
buf= 10000
• Computational complexity:
• Memory requirement:
Algorithm Complexity MIPS
QIC-GSC (3N-1)FFT + 14N – 12 1.38
SDW-MWF (no approximation) (3M+5)FFT + 28M + 6 3.46 SDW-MWF (approximations) (3M+2)FFT + 8M
2+ 14M + 3 2.8
Algorithm Memory kWords
QIC-GSC 4(N-1)L + 6L 0.64
SDW-MWF (no approximation)
2MLbuf+ 6LM + 7L 41.22 SDW-MWF (approximations)
4LM2+ 6LM + 7L 2.24
Complexity comparable to FD implementation of QIC-GSC
Substantial memory reduction through approximations
Introduction
Fixed beamforming
Adaptive stage
Implementation -Stochastic gradient -Complexity and
memory
Conclusions
26
Conclusions Conclusions
• Design of robust multimicrophone noise reduction algorithm:
o Design of robust fixed spatial preprocessor
need for statistical information about microphones o Design of robust adaptive stage
take speech distortion into account in cost function
• SP-SDW-MWF encompasses GSC and MWF as special cases
• Experimental results:
o SP-SDW-MWF achieves better noise reduction than QIC-GSC, for a given maximum speech distortion level
o Filter
w0improves performance in presence of model errors
• Implementations: Stochastic gradient algorithms available at affordable complexity and memory
• Further research: robustness against VAD-errors
o e.g. parameters dependent on input SNR
Spatially pre-processed SDW Multichannel Wiener Filter
Introduction
Fixed beamforming
Adaptive stage
Implementation
Conclusions
2727
Relevant publications Relevant publications
• Doclo S., Moonen M., “GSVD-Based Optimal Filtering for Single and Multi-Microphone Speech Enhancement”, IEEE Transactions on Signal Processing, vol. 50, no. 9, Sep.
2002, pp. 2230-2244.
• Spriet A., Moonen M., Wouters J., “A multichannel subband GSVD approach to speech enhancement”, European Transactions on Telecommunications, vol. 13, no. 2, Mar.
2002, pp. 149-158.
• Spriet A., Moonen M., Wouters J., “Robustness analysis of GSVD based optimal
filtering and generalized sidelobe canceller for hearing aid applications”, in Proc. of the IEEE Workshop on Applications on Signal Processing to Audio and Acoustics, New Paltz, New York, Oct. 2001, pp. 31-34.
• Maj J.B., Royackers L., Moonen M., Wouters J., “SVD-Based Optimal Filtering Technique For Noise Reduction In Dual Microphone Hearing Aids: A Real Time Evaluation”, submitted to IEEE Transactions on Biomedical Engineering.
• Doclo S., Moonen M., “Design of far-field and near-field broadband beamformers using eigenfilters”, Signal Processing, vol. 83, no. 12, pp. 2641-2673, Dec. 2003.
• Doclo S., Moonen M., “Design of broadband beamformers robust against gain and phase errors in the microphone array characteristics”, IEEE Transactions on Signal Processing, vol. 51, no. 10, Oct. 2003, pp. 2511-2526.
• Doclo S., Moonen M., “Design of broadband beamformers robust against microphone position errors”, in Proc. of the International Workshop on Acoustic Echo and Noise Control, Kyoto, Japan, Sep. 2003, pp. 267-270.
• Spriet A., Moonen M., Wouters J., “Spatially pre-processed speech distortion weighted multi-channel Wiener filtering for noise reduction”, Internal Report 03-46, ESAT-
SISTA, K.U.Leuven (Leuven, Belgium), 2003.
• Spriet A., Moonen M., Wouters J., “Stochastic Gradient based implementation of spatially pre-processed multi-channel Wiener filtering for noise reduction in hearing aids”, Internal Report 03-47, ESAT-SISTA, K.U.Leuven (Leuven, Belgium), 2003.
Available at SISTA publication engine:
http://www.esat.kuleuven.ac.be/~sistawww/cgi-bin/pub.pl
Introduction
Fixed beamforming
Adaptive stage
Implementation
Conclusions