Sinewave Amplitude Coding Using Line Spectral Frequencies
R. McAulay, Terrence G. Champion, Thomas F. Quatieri · 2005
The Sinusoidal Transform Coder (STC) is a vocoding technique that has demonstrated synthetic speech of good quality at rates from 2400 b/s to 4800 b/s [1]. Heretofore, a cepstral model has been fitted to the sinewave amplitudes, and the cepstral coefficients have been quantized by transforming to a set a channel gains that represent a non-integer downsampling of the spectral envelope [1]. This paper reports on a new approach that fits an all-pole model to the sinewave amplitudes and performs coding in terms of both scalar and vector quantization of the line spectral frequencies (LSFs). Good quality synthetic speech is obtained at 2400, 4800 and 8000 b/s with the use of 18, 22 and 32 LSFs. The use of a subband vector quantizer [8] improver, performance considerably at the 2400 b/s data rate.