Summary A new monaural method for the suppression of late room reverberation from speech signals, based on spectral subtraction, is presented. The problem of reverberation suppression differs from classical speech de-noising in that the “reverberation noise” is non stationary. In this paper, the use of a novel estimator of the non-stationary reverberationnoise power spectrum, based on a statistical model of late reverberation, is presented. The algorithm is tested on real reverberated signals. The performances for different RIRs with ranging from 0.34 s to 1.7 s consistently show significant noise reduction with little signal distortion. Moreover, when used as a front end to an automatic speech recognition system, the algorithm brings about dramatic improvements in terms of automatic speech recognition scores in various reverberant environments.

/pdf/a-new-method-based-on-spectral-subtraction-for-speech-4g7r5a110d.pdf

A New Method Based on Spectral Subtraction for Speech Dereverberation

Principal characteristics of speech speech production models speech analysis and analysis-synthesis systems linear predictive coding (LPC) analysis speech coding speech synthesis speech recognition future directions of speech processing. Appendices: convolution and z-transform vector quantization algorithm neural nests.

Digital speech processing, synthesis, and recognition

A directional acoustic receiving system is constructed in the form of a necklace including an array of two or more microphones mounted on a housing supported on the chest of a user by a conducting loop encircling the user's neck. Signal processing electronics contained in the same housing receives and combines the microphone signals in such a manner as to provide an amplified output signal which emphasizes sounds of interest arriving in a direction forward of the user. The amplified output signal drives the supporting conducting loop to produce a representative magnetic field. An electroacoustic transducer including a magnetic field pick-up coil for receiving the magnetic field is mounted in or on the user's ear and generates an acoustic signal representative of the sounds of interest. The microphone output signals are weighted (scaled) and combined to achieve desired spatial directivity responses. The weighting coefficients are determined by an optimization process. By bandpass filtering the weighted microphone signals with a set of filters covering the audio frequency range and summing the filtered signals, a receiving microphone array with a small aperture size is caused to have a directivity pattern that is essentially uniform over frequency in two or three dimensions. This method enables the design of highly-directive hearing instruments which are comfortable, inconspicuous, and convenient to use. The invention provides the user with a dramatic improvement in speech perception over existing hearing aid designs, particularly in the presence of background noise and reverberation.

A directional hearing system

We present an analysis of the generalized sidelobe canceller (GSC). It can be shown that the theoretical limits of the noise reduction performance depend only on the auto- and cross-spectral densities of the input signals. Furthermore, we compute the limits of the noise reduction performance for the theoretically determined diffuse noise field, which is an approximation for reverberant rooms. Our results show that the GSC cannot reduce noise further than 1 dB. These results were verified by simulation of reverberant environments. Only in sound-proof rooms with a reverberation time less than 100 ms the GSC performs well.

Theoretical noise reduction limits of the generalized sidelobe canceller (GSC) for speech enhancement

A system and method for a directional microphone system is disclosed. The directional microphone system can adaptively track and detect sources of sound information, and can reduce background noise. A first monolithic detection unit for detecting sound information and performing local signal processing on the detected sound information is provided. In the detection unit, an integrated transducer is provided for receiving acoustic waves and for generating sound information representative of the waves. A processor is coupled to the transducer for receiving the sound information and for performing local digital signal processing on the sound information to generate locally processed sound information. A base unit is coupled to the first monolithic detection unit and includes a global processor which receives the locally processed sound information and performs global digital signal processing on the locally processed sound information to generate globally processed sound information.

System and method for a monolithic directional microphone array

Speech Enhancement Using Microphone Array with Multi-Stage Processing

Voiced/Unvoiced/Silence Classification of Noisy Speech in Real Time Audio Signal Processing

An artificial neural network (ANN) speech-classifier-controlled iterative filtering system is described, which simulates the cocktail party effect for speech separation. The ANN speech classifier controls a modified iterative Wiener filter to cancel the interference by setting the filter's parameters and the convergence criterion for the iteration. The proposed system has been employed successfully with multiple-microphone speech acquisition systems for co-talker speech separation. The simulation results have shown that the iterative processing controlled by the neural network consistently provides speech of good quality and intelligibility.

Co-talker Separation Using the 'Cocktail Party Effect'

A multi-microphone speech enhancement system including two stages of processing has been developed. In the first stage, the array is aimed at the desired signal source and the noise signal source(s) respectively by means of the Frost algorithm or DS beamforming. In the second stage, the difference between the SNRs of the array outputs are employed to improve the SNR of the desired signal further by an adaptive filter with MRSS (matrix ratio sub space) algorithm. This algorithm is based on the orthogonal and coincidental subspace eigen analysis. The mathematical derivation and the simulation results that illustrate the advantages of the system are presented. >

http://ieeexplore.ieee.org/iel2/1116/8027/00344919.pdf

Post-microphone-array speech enhancement with adaptive filters for forensic application

This paper describes a novel speech enhancement structure which simulates the cocktail party effect using a modified iterative Wiener filter and a multi-layer perceptron neural network. The key idea is to use the neural network as a speaker recognition system to control the iterative Wiener filter. The neural network is a modified perceptron with a hidden layer using feature data extracted from LPC cepstral analysis. The proposed technique has been successfully used for speech enhancement when the interference is competing speech or broad band noise.

Yuchang Cao

Papers

Speech Enhancement Using Microphone Array with Multi-Stage Processing

Voiced/Unvoiced/Silence Classification of Noisy Speech in Real Time Audio Signal Processing

Co-talker Separation Using the 'Cocktail Party Effect'

Post-microphone-array speech enhancement with adaptive filters for forensic application

Speech Enhancement Iby Simulation Of Cocktail Party Effect With Neural Network Controlled Iterative Filter