Autoregressive Modelling for Speech Coding: Estimation, Interpolation andQuantisation (Autoregressieve Modellering voor Spraakcodering: Schatten, Interpoleren en Kwantiseren)

摘要

Linear Prediction has been applied very successfully in speech coding and manycoders are based on this technique. These coders use an autoregressive model to describe the speech signal. The reconstructed speech is synthesised by feeding a suitable excitation through the autoregressive synthesis filter. Coders differ mainly in the way the excitation through the autoregressive synthesis filter. Coders differ mainly in the way the excitation is selected and coded. Perhaps the main reason for the success of Linear Prediction is that the autoregressive model combines good prediction capabilities in the time domain with an accurate description of the speech spectral envelope. These properties can be used for efficient coding of the model and its excitations and for techniques such as interpolation, error weighting and postfiltering. This thesis focuses on different aspects of the application of autoregressive models for speech coding, such as estimation, interpolation and quantisation. (Copyright (C) 1996 J. S. Erkelens.)