This paper presents a robust algorithm for voice activity detection (VAD) based on newly established properties of the higher order statistics (HOS) of speech. Analytical expressions for the third and fourth-order cumulants of the LPC residual of short-term speech are derived assuming a sinusoidal model. The flat spectral feature of this residual results in distinct characteristics for these cumulants in terms of phase, periodicity and harmonic content and yields closed-form expressions for the skewness and kurtosis. Important properties about these cumulants and their similarity with the autocorrelation function are revealed from this exploratory part. They show that the HOS of speech are sufficiently distinct from those of Gaussian noise and can be used as a basis for speech detection. Their immunity to Gaussian noise makes them particularly useful in algorithms designed for low SNR environments. The proposed VAD algorithm combines HOS metrics with second-order measures, such as SNR and LPC prediction error, to classify speech and noise frames. The variance of the HOS estimators is quantified and used to yield a likelihood measure for noise frames. Moreover, a voicing condition for speech frames is derived based on the relation between the skewness and kurtosis of voiced speech. The performance of the algorithm is compared to the ITU-T G.729B VAD in various noise conditions, and quantified using the probability of correct and false classifications. The results show that the proposed algorithm has an overall better performance than G.729B, with noticeable improvement in Gaussian-like noises, such as street and parking garage, and moderate to low SNR.

Robust voice activity detection using higher-order statistics in the LPC residual domain

A robust algorithm for accurate endpointing of speech signals

A method for measurement of the fundamental frequency of a voiced speech signal corrupted by high levels of additive white Gaussian noise is described. The method is based on flattening the spectrum of the signal by a bank of bandpass lifters and extracting the pitch frequency from autocorrelation functions calculated at the output of the lifters. A smoothing modified median filter is applied to the calculated pitch frequency contour to result in an improvement in the accuracy of the method. A byproduct of the pitch tracker is a voiced/ unvoiced classifier. The maximum and the variance of the autocorrelation function maxima, over the bank of lifters, serve as the basis for voiced/unvoiced classification by making use of a two-dimensional, nearest-neighbor pattern recognition approach. Results are presented for fundamental frequency measurement and voiced/unvoiced classification for several signal-to-noise ratios.

A spectral autocorrelation method for measurement of the fundamental frequency of noise-corrupted speech

The results of designing, training, and testing a neural network for the voiced/unvoiced (V/UV) speech classification problem are described A feedforward multilayer backpropagation network was used with six input, ten internal, and two output nodes-for a binary decision The six features are common and easily computed Training was done with 72 frames from two speakers Testing was done with 479 frames from four speakers and resulted in a total of two errors (04%) Thus, a small neural network performs well on the V/UV problem >

Neural networks for voiced/unvoiced speech classification

Experimental results are presented that provide some insight into the higher-order statistical properties of speech signals. It is demonstrated that the third- and fourth-order distributions of the speech signal can be considered as mixtures of, rather than single, spherically invariant (Gaussian) distributions, where the mixing is controlled by the history of the process. >

On the higher order distributions of speech signals

A composite-Gaussian source model for speech was suggested at ICASSP-79. Based upon this model a voice/unvoiced detector is derived. This detector is an approximation to a maximum-a-posteriori sequence estimator and employs the Viterbi algorithm. This paper deals with the performance of this detector with real speech.

V. Ramamoorthy

Papers

Voice/Unvoice detection based on a composite-Gaussian source model of speech