Digital coding of speech using code excited linear prediction

Abstract

In this thesis the coding of narrow-band speech at rates between four and sixteen thousand bits per second has been studied. The work has concentrated on the Code Excited Linear Prediction (CELP) algorithm, and on means of improving this algorithm.

Conventional CELP coders-decoders (codecs) employing forward adaptive linear prediction analysis have been studied at bit rates between 4 and 8 kilobits per second (kbits/s). These CELP codecs offer quality reconstructed speech due to their Analysis-by-Synthesis (AbS) structure. This AbS structure has been investigated in detail and ways of extending the AbS loop, and hence improving the quality of the codecs, have been studied.

The recent move towards digital mobile communication systems, and the popularity of such systems, means that a very important aspect of many speech codecs is their sensitivity to bit errors between the encoder and decoder. Various methods of improving this sensitivity are investigated, and also a new method of measuring the sensitivity is proposed to allow channel and speech codecs to be finely matched.

Traditional forward adaptive CELP codecs have a buffering frame size of 20 or 30 ms, and a delay of the order of 70 ms. This delay can cause problems, especially over networks, and so recently a 16 kbits/s CELP codec with a delay of less than 2 ms has been standardised. The extension of this codec to operate as a variable rate codec from 16 down to 8 kbits/s has been studied. Also other low delay CELP codecs operating at bit rates as low as 4 kbits/s with delays of less than 10 ms have been proposed and studied.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata