Emanuël A. P. Habets, Jacob Benesty, and Patrick A. Naylor Presented by Amir Kiperwas 1
M-element microphone array One desired source One undesired source Ambient noise field Signals: Broadband Mutually independent Zero mean Noise - Sufficiently stationary 2
RTF- relative transfer function online estimation methods available [11], [15] 3
Beamformer output : Beamformer weight vector 4
Of a random variable: Of a column vector b(w): Of the received signal: Of the Beamformer output: 5
Constraints: Z D U v 1 1 h H 6
SDIRC Beamformer: d 1 a u v 1 a SDIRC v h h h 1 2 1 H 1 h h A A A 1 2 v v 1 7
LCMV linearly constrained minimum variance (Frost) Try to extract the desired signal coming from a specific direction while minimizing contributions to the output due to interfering signals and noise arriving from directions other than the direction of interest. 8
MVDR minimum variance distortionless response (Capon) perhaps the most widely used adaptive beamformer minimize the output power with the constraint that the desired signal is not affected. MVDR 1 9
MVDR finding value for 10
PMWF parameterized multichannel Wiener filter. derived from the classical MSE criterion. limited in practice because of its lack of flexibility. we can not control the compromise between noise reduction and speech distortion 11
PMWF parameterized multichannel Wiener filter. Proportional to the Capon filter (MVDR) 12
(Narrowband) Input Signal-to-Noise Ratio Input Signal-to-interference Ratio Input Signal-to-interference plus noise Ratio Input Noise -to-interference Ratio (Fullband) 13
(Narrowband) Output Signal-to-Noise Ratio Output Signal-to-interference Ratio Output Signal-to-interference plus noise Ratio Output Noise -to-interference Ratio 14
Scenario I: One undesired source White ambient noise 15
Scenario I: osinr h ; osinr h osinr h PMWF MVDR LCMV 16
(Narrowband) Noise Reduction - quantifies the amount of noise being rejected by the beamformer. defined as the ratio of the power of the noise at the reference microphone over the power of the noise remaining at the beamformer output. 17
Scenario I: 18
(Narrowband) Speech Distortion- desired-signal-cancellation factor. the ratio of the variance of the desired signal at the reference microphone over the variance of the filtered desired signal at the beamformer output. Speech-distortion-index 19
(Narrowband) the ratio of the output SINR (after beamforming) over the input SINR (at the reference microphone). equal to the SINR improvement. 20
Simulation I: anechoic environment four microphones and an inter-microphone distance of 2.5 cm. Simulation II: reverberant environment. Room size = 5x4x6m The A(w) vector is computed using: 21
Processing: STFT, 50% overlap, Tframe = 512ms. 22
SIR = 10dB SNR =20dB 23
SIR = -5dB SNR =20dB 24
SIR = 10dB, SNR =20dB 25
anechoic chamber at Bell Labs rectangular room (6.7m x 6.1m x 2.9m). reverberation time of approximately 130 ms. SIR = 5dB, SNR =15dB Desired source - female speaker active between 0 and 2 s and between 3.5 and 5 s undesired source - male speaker active between 2 and 5s ambient noise - spatially homogeneous and spatially and temporally white Gaussian. 26
Desired signal at the first microphone 27
First microphone signal 28
Processed signal with 0, 0 LCMV 29
Processed signal with 0, 1 MVDR 30
Processed signal with 0, 0.5 31
Processed signal with 1, 0.5 32
33
34
Trade off: High speech distortion + High interference-plus-noise reduction 0 Low speech distortion + Low interference-plus-noise reduction 0 1 High undesired signal reduction High ambient noise reduction 35
Try to get a better understanding of the influence of and on the signal intelligibility using Several objective speech quality measures. PESQ (Perceptual Evaluation of Speech Quality). LPC-based (LLR, IS) Time-domain and frequency-weighted SNR measures. (fwsegsnr) 36
37
[1] J. Benesty, J. Chen, and Y. Huang, Microphone Array Signal Processing. Berlin, Germany: Springer-Verlag, 2008. [11] S. Gannot, D. Burshtein, and E.Weinstein, Signal enhancement using beamforming and nonstationarity with applications to speech, IEEE Trans. Signal Process., vol. 49, no. 8, pp. 1614 1626, Aug. 2001. [15] I. Cohen, Relative transfer function identification using speech signals, IEEE Trans. Speech Audio Process., vol. 12, no. 5, pp. 451 459, Sep. 2004. [16] Hu, Y. and Loizou, P. (2008). Evaluation of objective quality measures for speech enhancement, IEEE Transactions on Speech and Audio Processing, 16(1), 229-238. [17] Ma, J., Hu, Y. and Loizou, P. (2009). "Objective measures for predicting speech intelligibility in noisy conditions based on new band-importance functions", Journal of the Acoustical Society of America, 125(5), 3387-3405. 38