Relative Transfer Function Identification Using Speech Signals

Abstract

Abstract-An important component of a multichannel hands-free communication system is the identification of the relative transfer function between sensors in response to a desired source signal. In this paper, a robust system identification approach adapted to speech signals is proposed. A weighted least-squares optimization criterion is introduced, which considers the uncertainty of the desired signal presence in the observed signals. An asymptotically unbiased estimate for the system's transfer function is derived, and a corresponding recursive online implementation is presented. We show that compared to a competing nonstationarity-based method, a smaller error variance is achieved and generally shorter observation intervals are required. Furthermore, in the case of a time-varying system, faster convergence and higher reliability of the system identification are obtained by using the proposed method than by using the nonstationarity-based method. Evaluation of the proposed system identification approach is performed under various noise conditions, including simulated stationary and nonstationary white Gaussian noise, and car interior noise in real pseudo-stationary and nonstationary environments. The experimental results confirm the advantages of proposed approach.

Phase reference for the generalized multichannel Wiener filter

“…Since in general the microphone signals contain correlated noise components, estimating the RTFs directly from the noisy microphone signals leads to biased RTF estimates. Several methods for unbiased RTF estimation have been proposed, e.g., by exploiting the non-stationarity of speech signals [13,33] or based on the generalized eigenvalue decomposition of R Y and R N [34,35]. In [36], an approach for unbiased RTF estimation was proposed, requiring estimates of the PSDs and CPSDs of the speech and noise components, which can be obtained from the estimated speech and noise correlation matrices R S and R N .…”

Sound source separation by using matched beamforming and time-frequency masking

“…6 shows the waveforms of the input and the matched beamformer output signals with respect to the RTF estimation method. We compared three methods, namely the BLS [3], the LMS [4], and the proposed SLS method. In this experiment, both the BLS and SLS methods initialize the estimated RTF when the source moves.…”

“…This would affect the sound source separation process which would eventually degrade the overall performance. An adaptive form of a least squares solution using the least mean squares (LMS) was introduced in previous studies [4]. However, this led to a significant increase in computational load.…”

Successive relative transfer function identification using single microphone speech enhancement

“…Gannot et al [5] exploit the nonstationarity of speech signals to estimate the RTF. Cohen [6] utilized the speech presence probability (SPP) in the time-frequency domain to identify the time-frequency instances that consist of speech signal. The time-frequency instances that consist of speech signal are then utilized to derive an RTF estimator.…”