First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

A Highly Programmable Formant Speech Synthesiser Utilising Parallel Processors

K. M. Curtis, G. M. Asher, S. E. Pack, J. Andrews

Department of Electrical and Electronic Engineering, University of Nottingham, UK

Speech synthesisers capable of reproducing high quality male speech from parametric information already exist. The parallel formant synthesiser developed by the Joint Speech Research Unit is recognised as one of the best of these with a real-time, five formant digital signal processor implementation being commercially available. Current research into copy synthesis of female speech shows that the inclusion of further formants is required for the synthesis of realistic female speech. The inherently sequential nature of a digital signal processor implementation limits this extension to the parallel formant synthesiser. The advent of parallel processors and the availability of the Transputer however makes possible a truly parallel highly versatile implementation of a real-time synthesiser, capable of multi-gender speech synthesis. This paper describes the development of such a synthesiser having seven formant resonators. The design of the synthesiser utilised a new technique to overcome the communications "bottleneck" associated with the Transputer due to it only having four links. The lack of Transputer links would have proved problematic when performing the resonator data communications and during the passing of the control parameters for each of the resonators of the parallel formant synthesis model. This optimisation technique is also briefly described.

Full Paper

Bibliographic reference.  Curtis, K. M. / Asher, G. M. / Pack, S. E. / Andrews, J. (1990): "A highly programmable formant speech synthesiser utilising parallel processors", In ICSLP-1990, 817-820.