It has been shown that the principal subspace based multi-channel Wiener filter (MWF) provides better performance than the conventional MWF for the interference suppression in the case of a single target source. It can efficiently estimate the target component in the principal subspace and the acoustic transfer function up to a scaling factor. However, as input signal-to-interference ratio (SIR) becomes lower, larger errors are incurred in the estimation of the acoustic transfer function by the principal subspace degrading the performance in interference suppression. In order to alleviate this problem, a principal subspace modification method was proposed in previous work, which uses a priori information on the direction of the target signal. In this paper, a frequency-band dependent interpolation technique is further employed for the principal subspace modification and the speech recognition test is conducted to demonstrate the practical usefulness of the proposed methods as a front processing for the speech recognizer in interferer-present environment.

Existing speech enhancement algorithms can improve speech quality but not speech intelligibility, and the reasons for that are unclear. In the present paper, we present a theoretical framework that can be used to analyze potential factors that can influence the intelligibility of processed speech. More specifically, this framework focuses on the fine-grain analysis of the distortions introduced by speech enhancement algorithms. It is hypothesized that if these distortions are properly controlled, then large gains in intelligibility can be achieved. To test this hypothesis, intelligibility tests are conducted with human listeners in which we present processed speech with controlled speech distortions. The aim of these tests is to assess the perceptual effect of the various distortions that can be introduced by speech enhancement algorithms on speech intelligibility. Results with three different enhancement algorithms indicated that certain distortions are more detrimental to speech intelligibility degradation than others. When these distortions were properly controlled, however, large gains in intelligibility were obtained by human listeners, even by spectral-subtractive algorithms which are known to degrade speech quality and intelligibility.

While most speech enhancement algorithms improve speech quality, they do not improve speech intelligibility in noise. The reasons for that remain unclear. In this paper, we present a theoretical framework that can be used to analyze potential factors influencing the intelligibility of processed speech. It is hypothesized that if distortions are properly controlled, then large gains in intelligibility can be achieved. To assess the perceptual effect of the various distortions that can be introduced by speech enhancement algorithms, intelligibility tests are conducted with human listeners. The results indicated that certain distortions are more critical than others. The result of listening tests suggested that when these distortions are properly controlled, substantial gains in intelligibility can be obtained.

It has been shown that large gains in speech intelligibility can be obtained by using the binary mask approach which retains the time-frequency (T-F) units of the mixture signal that are stronger than the interfering noise (masker) (i.e., SNR>0 dB), and removes the T-F units where the interfering noise dominates. In this paper, we introduce a new binary mask for improving speech intelligibility based on noise distortion constraints. A binary mask is designed to retain noise overestimated T-F units while discarding noise underestimated T-F units. Listening tests were conducted to evaluate the new binary mask in terms of intelligibility. Results from the listening tests indicated that large gains in intelligibility can be achieved by the application of the proposed binary mask to noise-corrupted speech even at extremely low SNR levels (-10 dB). Index Terms: speech intelligibility, noise estimation, speech enhancement

A new binary mask is introduced for improving speech intelligibility based on magnitude spectrum constraints. The proposed binary mask is designed to retain time-frequency (T-F) units of the mixture signal satisfying a magnitude constraint while discarding T-F units violating the constraint. Motivated by prior intelligibility studies of speech synthesized using the ideal binary mask, an algorithm is proposed that decomposes the input signal into T-F units and makes binary decisions, based on a Bayesian classifier, as to whether each T-F unit satisfies the magnitude constraint or not. Speech corrupted at low signal-to-noise (SNR) levels (-5 and 0 dB) using different types of maskers is synthesized by this algorithm and presented to normal-hearing listeners for identification. Results indicated substantial improvements in intelligibility over that attained by human listeners with unprocessed stimuli.

While most speech enhancement algorithms improve speech quality, they may not improve speech intelligibility in noise. This paper focuses on the development of an algorithm that can be optimized for a specific acoustic environment and improve speech intelligibility. The proposed method decomposes the input signal into time-frequency (T-F) units and makes binary decisions, based on a Bayesian classifier, as to whether each T-F unit is dominated by the target signal or the noise masker. Target-dominated T-F units are retained while masker-dominated T-F units are discarded. The Bayesian classifier is trained for each acoustic environment using an incremental approach that continuously updates the model parameters as more data become available. Listening experiments were conducted to assess the intelligibility of speech synthesized using the incrementally adapted models as a function of the number of training sentences. Results indicated substantial improvements in intelligibility (over 60% in babble at -5 dB SNR) with as few as ten training sentences in babble and at least 80 sentences in other noisy conditions.

The ideal binary mask, often used in robust speech recognition applications, requires an estimate of the local SNR in each timefrequency (T-F) unit. A data-driven approach is proposed for estimating the instantaneous SNR of each T-F unit. By assuming that the a priori SNR and a posteriori SNR are uniformly distributed within a small region, the instantaneous SNR is estimated by minimizing the localized Bayes risk. The binary mask estimator derived by the proposed approach is evaluated in terms of hit and false alarm rates. Compared to the binary mask estimator that uses the decision-directed approach to compute the SNR, the proposed data-driven approach yielded substantial improvements (up to 40%) in classification performance, when assessed in terms of a sensitivity metric which is based on the difference between the hit and false alarm rates. Index Terms: ideal binary mask, SNR estimation, Bayes risk 1.

In multi-microphone noise reduction for single desired speech signal, the principal subspace based multi-channel Wiener filter provides better performance compared with the conventional multi-channel Wiener filter. The principal subspace vector estimates the acoustic transfer function vector up to a scaling factor. However, as input SNR becomes lower, the error increases in the acoustic transfer function vector estimation. In this paper, we propose the principal subspace modification which is controlled by the angle between the principal subspace vector and the steering vector of the desired speech signal. In the simulation, the proposed method is evaluated with multi-channel speech data which are degraded by interfering noise coming from other direction. The simulation results show that the modification of principal subspace vector allows better performance compared to the conventional principal subspace based multi-channel Wiener filter.

In this paper, frequency domain multi-channel filtering schemes are proposed for speech enhancement, based on the subspace decomposition of spatial spectral matrices. For better estimation of noise statistics, which is important for most speech enhancers, we propose noise eigenvalue modification methods for the correction of noise spatial spectral matrix. These methods are based on the rank-1 property of the speech spatial spectral matrix for single desired speech source. Simulation results show that the proposed methods yield better performance compared to the conventional multi-channel Wiener filtering.