Breen, Andrew Paul; (1991) An investigation into synthesis-by-analysis. Doctoral thesis (Ph.D), UCL (University College London).

Text An_investigation_into_synthesi.pdf Download (8MB)

Abstract

The problem addressed in this thesis was the design of a fast, automatic method of synthesis-by-analysis, which did not compromise the quality of the synthesis for speed of analysis. The technique was implemented in software and used the JSRU parallel formant synthesiser. The thesis begins by placing the subject of speech synthesis in historical perspective with an overview of speech production theories and their incorporation in past and present day synthesiser designs. The technique of synthesis-by-analysis is then described and attention drawn to its many applications. As the new method of synthesis-by-analysis requires good voicing, pitch and formant frequency control signal values, standard methods for determining these control signals are described, with particular attention paid to methods which employ electrolaryngographic techniques. A new method of formant amplitude estimation is introduced termed 'automatic amplitude mapping' (AAM), which tailors the extraction of amplitude estimates to suit the design requirements of the synthesiser used. To improve on these initial estimates, the amplitude values derived from AAM are processed using neural networks to produce optimised formant synthesiser control signals. This new technique has been termed 'transform synthesis'. Experiments are described which use subjective measures to compare the quality of synthetic speech produced using AAM and transformed AAM values, with a reference synthesis-by-analysis method. Results are given for a number of statistical analyses conducted using data obtained from these experiments. Results are also presented for experiments which investigated the effect of prosody, speaker sex, and style on the glottal closed quotient. Initial results are given for experiments which attempted to incorporate such excitation specific information into the JSRU synthesiser using a fixed voiced excitation and a dynamically varying voiced excitation.

Download activity - last month

Export as XMLCSVJSON

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item