Speech coding

About: Speech coding is a research topic. Over the lifetime, 14245 publications have been published within this topic receiving 271964 citations.

Adaptive transform coding of speech signals

Abstract: This paper discusses speech coding systems based upon transform coding (TC). It compares several transforms and shows that the cosine transform leads to a nearly optimum performance for almost all speech sounds. Various adaptive coding strategies are then investigated, and a coding scheme is proposed that is based on a nonadaptive discrete cosine transform (DCT), on an adaptive bit assignment, and on adaptive quantization. The adaptation is controlled by a short-term basis spectrum that is derived from the transform coefficients prior to coding and transmission and that is transmitted as side information to the receiver. The main result is that this adaptive transform coder performs better than all known nonpitch-tracking coding schemes; it extends the range of speech waveform coding to lower bit rates and closes the gap between vocoders and predictive waveform coders.

Estimation of Glottal Closure Instants in Voiced Speech Using the DYPSA Algorithm

Abstract: We present the Dynamic Programming Projected Phase-Slope Algorithm (DYPSA) for automatic estimation of glottal closure instants (GCIs) in voiced speech. Accurate estimation of GCIs is an important tool that can be applied to a wide range of speech processing tasks including speech analysis, synthesis and coding. DYPSA is automatic and operates using the speech signal alone without the need for an EGG signal. The algorithm employs the phase-slope function and a novel phase-slope projection technique for estimating GCI candidates from the speech signal. The most likely candidates are then selected using a dynamic programming technique to minimize a cost function that we define. We review and evaluate three existing methods of GCI estimation and compare the new DYPSA algorithm to them. Results are presented for the APLAWD and SAM databases for which 95.7% and 93.1% of GCIs are correctly identified

Experimental evaluation of features for robust speaker identification

Abstract: This correspondence presents an experimental evaluation of different features and channel compensation techniques for robust speaker identification. The goal is to keep all processing and classification steps constant and to vary only the features and compensations used to allow a controlled comparison. A general, maximum-likelihood classifier based on Gaussian mixture densities is used as the classifier, and experiments are conducted on the King speech database, a conversational, telephone-speech database. The features examined are mel-frequency and linear-frequency filterbank cepstral coefficients, linear prediction cepstral coefficients, and perceptual linear prediction (PLP) cepstral coefficients. The channel compensation techniques examined are cepstral mean removal, RASTA processing, and a quadratic trend removal technique. It is shown for this database that performance differences between the basic features is small, and the major gains are due to the channel compensation techniques. The best "across-the-divide" recognition accuracy of 92% is obtained for both high-order LPC features and band-limited filterbank features. >

Sparse Autoencoder-Based Feature Transfer Learning for Speech Emotion Recognition

Abstract: In speech emotion recognition, training and test data used for system development usually tend to fit each other perfectly, but further 'similar' data may be available. Transfer learning helps to exploit such similar data for training despite the inherent dissimilarities in order to boost a recogniser's performance. In this context, this paper presents a sparse auto encoder method for feature transfer learning for speech emotion recognition. In our proposed method, a common emotion-specific mapping rule is learnt from a small set of labelled data in a target domain. Then, newly reconstructed data are obtained by applying this rule on the emotion-specific data in a different domain. The experimental results evaluated on six standard databases show that our approach significantly improves the performance relative to learning each source domain independently.

ITU-T Recommendation G.729 Annex B: a silence compression scheme for use with G.729 optimized for V.70 digital simultaneous voice and data applications

Abstract: This article describes Annex B to ITU-T Recommendation G.729. Annex B defines a low-bit-rate silence compression scheme designed and optimized to work in conjunction with both the full version of G.729 and its low-complexity Annex A. To achieve good quality low-bit-rate silence compression, a robust frame-based voice activity detector module is essential to detect inactive voice frames, also called silence or background noise frames. For these detected inactive voice frames, a discontinuous transmission module measures the changes over time of the inactive voice signal characteristics and decides whether a new silence information descriptor frame should be sent to maintain the reproduction quality of the background noise at the receiving end. If such a frame is needed, the spectrum and energy parameters describing the perceptual characteristics of the background noise are efficiently coded and transmitted using 15 b/frame. At the receiving end, the comfort noise generation module regenerates the output background noise using transmitted updates or previously available parameters. The synthesized background noise is obtained by linear predictive filtering of a locally generated pseudo-white excitation signal of a controlled level. This method of coding the background noise enables the achievement of bit-rate savings for coded speech at average rates as low as 4 kb/s during normal speech conversation while maintaining reproduction quality.