Title :
A spline approach to speech analysis/Synthesis
Author_Institution :
University of Utah, Salt Lake City, Utah
Abstract :
Vocoders based on linear models of the vocal tract such as LPC result in an inverse polynomial fit of the spectrum, and along with homomorphic vocoders require pitch estimation. Analysis/synthesis using B-spline basis functions is proposed in this preliminary paper. This is a local approximation scheme which permits a concentration of parameters in regions of greater importance and employs easily computed least square coefficients. It can be used with pitch based vocoders or as a standalone nonpitch based vocoder. An experimental system, not yet optimized using special properties of human speech, has been applied to samples of male speech, female speech, simultaneous speech with two speakers, and noisy speech. Empirical results on tape will be presented.
Keywords :
Convolution; Equations; Fourier transforms; Least squares approximation; Least squares methods; Low pass filters; Polynomials; Speech analysis; Speech synthesis; Spline;
Conference_Titel :
Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP '81.
DOI :
10.1109/ICASSP.1981.1171268