Siow Yong Low

Convolutive blind signal separation with post-processing

A new subband based speech enhancement scheme is presented. It integrates spatial and temporal signal processing methods to enhance speech signals in a noisy environment. The approach makes use of the popular blind signal separation (BSS) to spatially separate the target signal from the interference. Due to the multipath/reverberant environment, BSS has its fundamental limitation in its separation quality. To overcome that, an adaptive noise canceller (ANC) is employed to perform further interference reduction. The reference for the ANC in this case is simply the interference dominant output from the BSS. A higher order statistical method is proposed for the selection of the reference signal. This post processing acts as a spectral decorrelator and experimental results show that even in under-determined (more sources than elements) case, the structure offers impressive enhancement capability. Further, a remarkable improvement in recognition rate is registered when tested in automatic speech recognition (ASR).

A subband space constrained beamformer incorporating voice activity detection [speech enhancement applications]

This paper introduces a new subband adaptive space constrained beamforming structure for use in hands-free speech enhancement applications. The scheme incorporates a space constrained source model and voice activity information through the integration of a voice activity detector (VAD). The VAD information is used to estimate noise covariance information during non-speech periods and to optimally estimate the source power spectral density (PSD), which is used to provide a spectrally optimized constraint on the source. The proposed structure is evaluated in a real car environment, yielding results which compare well to the optimal Wiener solution where full knowledge of the source is known.

Blind Signal Separation Using Steepest Descent Method

A method that significantly improves the convergence rate of the gradient-based blind signal separation (BSS) algorithm for convolutive mixtures is proposed. The proposed approach is based on the steepest descent algorithm suitable for constrained BSS problems, where the constraints are included to ease the permutation effects associated with the convolutive mixtures. In addition, the method is realized using a modified golden search method plus parabolic interpolation, and this allows the optimum step size to be determined with only a few calculations of the cost function. Evaluation of the proposed procedure in simulated environments and in a real room environment shows that the proposed method results in significantly faster convergence for the BSS when compared with a fixed step-size gradient-based algorithm. In addition, for blind signal extraction where only a main speech source is desired, a combined scheme consisting of the proposed BSS and a postprocessor, such as an adaptive noise canceller, offers impressive noise suppression levels while maintaining low-target signal distortion levels.

Space constrained beamforming with source PSD updates

This paper presents a new space constrained adaptive beamformer employing an updated source power spectral density (PSD). The space constraints are used to capture the target signal spatially and to provide robustness against steering error vectors. The PSD update on the other hand ensures that the spectral information of the desired source is reflected continuously on the space constraints. As such, target signal extraction can be achieved with minimum distortion. The beamformer operates in a subband structure to allow time-frequency operation for each channel, yielding a combination of weighted spatial and temporal filters. Evaluations on real car data show that the proposed algorithm significantly improves the speech intelligibility with noise suppression level up to 21 dB.

Spatio-temporal processing for distant speech recognition

A new subband based front-end processor for speech recognition is presented. It integrates both spatial and temporal signal processing methods to enhance noisy signals as a means to reduce the mismatch problem in speech recognition. The approach makes use of the popular blind signal separation (BSS) to spatially separate the target signal from the interference. Due to the multipath/reverberant environment, BSS has its fundamental limitation in the separation quality. To overcome that, an adaptive noise canceller (ANC) is employed to perform further interference reduction. Experimental results show that even in an adverse environment, the proposed structure improves the word recognition rate (WRR) by 70% for the connected digit recognition task.

Adaptive microphone array with noise statistics updates

In this paper, we present a new subband adaptive beamformer equipped with noise statistics updates. These updates are employed to effectively estimate and track the noise statistics continuously in the solution. Additionally, an update on the source power spectral density (PSD) is incorporated to enhance the timbre of the source of interest. Furthermore, the beamformer is also equipped with a space constraint on the source area to provide robustness against steering vector errors and good capture of the target signal spatially. The entire processing can be viewed as an efficient combination of weighted spatial and temporal filters. Evaluations on real car data with variations in the car noise level show that the proposed scheme achieves a good noise suppression level up to 20 dB.

Speech Signal Extraction Utilizing PCA-ICA Algorithm With a Non-Uniform Spacing Microphone Array

Speech signal extraction is becoming more and more important as evidently displayed by its numerous applications such as mobile phones, conference equipments and surveillance. This paper presents a blind method to enhance a speech source of interest in noisy environments. The proposed technique consists of the principal component analysis (PCA) and the independent component analysis (ICA) to extract the speech signal. In an effort to overcome the small phase resolution due to the constraint on the inter-element distance, a non-uniform spacing PCA-ICA algorithm is suggested. By utilizing a different inter-element distance processing on each pair of microphones in a multistage fashion, a better separation is achieved. Results show better separation performance for the proposed method compared to the uniformly spaced microphone array

A blind approach to joint noise and acoustic echo cancellation

The paper introduces a new scheme which combines the popular blind signal separation (BSS) and a post-processor to suppress noise and acoustic echo jointly. The new L element structure uses the BSS as a front-end processor to extract the target signal spatially from the interference (noise and echo). Statistical measures are then employed to select the target signal dominant signal from the BSS outputs. The remaining L-1 BSS outputs (noise and echo dominant) and the existing far-end line echo are then used as the reference signals in an adaptive noise canceller (ANC) to enhance the target signal temporally. The novel structure bypasses the need for any a priori information whilst compensating the separation quality of the BSS temporally. Real room evaluations demonstrate the efficacy of the scheme in both noisy double-talk and non double-talk situations.

Speech enhancement using multiple soft constrained subband beamformers and non-coherent technique

This paper presents a new robust microphone array processing technique to enhance speech signals under the influence of noise and jammer(s). The new structure comprises two soft constrained subband beamformers and a non-coherent processing technique. Essentially, the first beamformer enhances the desired speech signal in a specified constrained region. The residual interference in the beamformers output is then spectral subtracted using the estimated interference from the second beamformer. Evaluations in a real office environment show higher interference suppression compared to those obtained using the soft constrained beamformer only. Most importantly, this is achieved with negligible expense on target signal distortion.

Postfiltering using multichannel spectral estimation in multispeaker environments

This paper investigates the problem of enhancing a single desired speech source from a mixture of signals in multispeaker environments. A beamformer structure is proposed which combines a fixed beamformer with postfiltering. In the first stage, the fixed multiobjective optimal beamformer is designed to spatially extract the desired source by suppressing all other undesired sources. In the second stage, a multichannel power spectral estimator is proposed and incorporated in the postfilter, thus enabling further suppression capability. The combined scheme exploits both spatial and spectral characteristics of the signals. Two new multichannel spectral estimation methods are proposed for the postfiltering using, respectively, inner product and joint diagonalization. Evaluations using recordings from a real-room environment show that the proposed beamformer offers a good interference suppression level whilst maintaining a low-distortion level of the desired source.