Sebastian Stenzel

Microphone diversity combining for in-car applications

This paper proposes a frequency domain diversity approach for two or more microphone signals, for example, for in-car applications. The microphones should be positioned separately to insure diverse signal conditions and incoherent recording of noise. This enables a better compromise for the microphone position with respect to different speaker sizes and noise sources. This work proposes a two-stage approach. In the first stage, the microphone signals are weighted with respect to their signal-to-noise ratio and then summed similar to maximum ratio combining. The combined signal is then used as a reference for a frequency domain least-mean-squares (LMS) filter for each input signal. The output SNR is significantly improved compared to coherence-based noise reduction systems, even if one microphone is heavily corrupted by noise.

A noise PSD and cross-PSD estimation for two-microphone speech enhancement systems

In this paper, we propose an efficient noise power and noise cross-power spectral density estimation method for distributed microphones. The proposal can be combined with coherence based two-microphone noise reduction systems. It utilizes a minimum statistic based noise PSD estimator for each channel and a joint voice activity detector. Evaluation results show that the voice activity detection significantly improves the individual estimates.

Blind-Matched Filtering for Speech Enhancement with Distributed Microphones

A multichannel noise reduction and equalization approach for distributed microphones is presented. The speech enhancement is based on a blind-matched filtering algorithm that combines the microphone signals such that the output SNR is maximized. The algorithm is developed for spatially uncorrelated but nonuniform noise fields, that is, the noise signals at the different microphones are uncorrelated, but the noise power spectral densities can vary. However, no assumptions on the array geometry are made. The proposed method will be compared to the speech distortion-weighted multichannel Wiener filter (SDW-MWF). Similar to the SDW-MWF, the new algorithm requires only estimates of the input signal to noise ratios and the input cross-correlations. Hence, no explicit channel knowledge is necessary. A new version of the SDW-MWF for spatially uncorrelated noise is developed which has a reduced computational complexity, because matrix inversions can be omitted. The presented blind-matched filtering approach is similar to this SDW-MWF for spatially uncorrelated noise but additionally achieves some improvements in the speech quality due to a partial equalization of the acoustic system.

Time-frequency dependent voice activity detection based on a simple threshold test

This paper proposes an efficient method to determine for each time-frequency point whether speech is present or absent in a noisy microphone signal. Based on a likelihood ratio test for the conditional speech presence probability a simple threshold test is derived. This test compares the current input power for each time-frequency point with an estimate of the noise power spectral density. The theoretical results as well as the simulation results indicate that voice activity is well approximated.

A multichannel Wiener filter with partial equalization for distributed microphones

In speech enhancement applications, the multichannel Wiener filter (MWF) is widely used to reduce noise and thus improve signal quality. The MWF performs noise reduction by estimating the desired signal component in one of the microphones, referred to as the reference microphone. However, for distributed microphones, the selection of the reference microphone has a significant impact on the broadband output SNR of the MWF, largely depending on the acoustical transfer function (ATF) between the desired source and the reference microphone. In this paper, a multichannel Wiener filtering approach using a soft combined reference is presented. Simulation results show that the proposed scheme leads to a higher broadband output SNR compared to an arbitrarily selected reference microphone, moreover achieving a partial equalization of the overall acoustic system.

An FLMS based two-microphone speech enhancement system for in-car applications

Distributed microphone arrays enable a better compromise for the microphone position with respect to different speaker sizes. They also ensure robust noise reduction with respect to different noise sources. This paper proposes a frequency domain LMS (FLMS) based algorithm to combine the signals of two microphones. It is shown that noise correlation leads to a bias of the FLMS filter transfer function. To overcome this issue a noise cross-power estimator for estimating and correcting the filter bias is proposed.

Alternative formulation and robustness analysis of the multichannel wiener filter for spatially distributed microphones

The multichannel Wiener filter (MWF) is a well-known multi-microphone noise reduction technique, which aims to estimate the speech component in one of the microphone signals. Assuming a single speech source, the rank-one property of the speech correlation matrix can be exploited to derive the so-called rank-one MWF (R1-MWF). In this paper, we present an alternative formulation of the MWF (A-MWF), which exploits the assumed rank-one property of the speech correlation matrix in a different way as the R1-MWF. Furthermore, we present a theoretical robustness analysis of the different MWF formulations in presence of spatially white noise. Experimental results show that similarly to the R1-MWF, the proposed A-MWF is less sensitive to estimation errors of the speech correlation matrix and yields a higher output SNR than the standard MWF.

A Minimum variance beamformer for spatially distributed microphones using a soft reference selection

The signal quality at different microphone locations varies for distributed microphone arrays. Such microphone constellations require a suitable beamformer design that considers these differences in the input signal conditions. In this paper a frequency domain minimum variance (MV) beamforming approach is presented that uses a soft reference selection. The filter coefficients of this MV beamformer only depend on the spatial correlation properties of the input signals. The proposed beamformer is able to partially equalize the acoustic system, i.e. the magnitude of the overall transfer function is the envelope of all acoustic transfer functions.

Time-frequency masking for convolutive and noisy mixtures

Time-frequency masking is a suitable tool for speech enhancement and source separation of speech signals. This paper presents an efficient method to determine the time-frequency masks for two speech signals based on two microphone signals in a noisy and reverberant environment. A two-stage processing is proposed. In the first stage two linear filters are adapted using the LMS algorithm. In the second stage the time-frequency masks are determined based on the transfer functions of the adaptive filters. The presented simulation results verify that the time-frequency masks are well approximated.

Combined echo and noise reduction for distributed microphones

This work presents an echo and noise suppression approach that can be used in teleconference systems. The problem in such hands-free telecommunication scenarios is the possible large distance between the local speakers and the microphones. The microphones are placed more than 40cm apart from each other to accomplish a better compromise for different speaker positions. A microphone diversity approach is presented which exploits these different signal and noise conditions at the microphones. In general teleconference systems require additional signal processing for echo cancellation. To reduce possible background noise from the microphone input signals the system contains a Multichannel Wiener Filter. The main contribution of this paper is an extension of the Multichannel Wiener Filter for combined noise reduction and residual echo suppression.