Speech Pre-Processing for Pitch and Pitch-Cylce Evolutions Smoothing

Authors: HASSAN FARSI

Abstract: In low bit rate speech coders, pitch is usually transmitted once per frame and, when needed, the intermediate pitch values are obtained by interpolation between 2 adjacent pitch values. Although pitch usually evolves slowly, sometimes it has irregular variations and the estimated pitch differs from the real one. In addition, some speech coders, e.g., waveform interpolation coders, rely on smooth pitch-cycle evolutions to extract speech model parameters in the analysis stage. However, non-stationary characteristics of speech may lead to inaccurate estimation of the parameters. This affects the synthesised speech quality. We propose a pre-processor, which modifies the residual speech signal to provide smooth pitch variations and pitch-cycle evolutions, without distorting perceptual speech quality. Thus, the pitch and the voicing level can be more accurately determined.

Keywords: Pitch, speech coder, bit rate, pitch cycle

Full Text: PDF