Pitch modification of speech signal using harmonic model with time-varying parameters

The paper presents a solution to the problem of pitch modification of speech. The problem occurs in different speech processing applications such as voice conversion, accent correction, hiding speaker's personality and other. Developed processing scheme for voiced part of speech is based on the harmonic model with nonstationary (time-varying) parameters. In order to improve frequency resolution of the model parameters are extracted using narrow-band filtering in warped time domain aligned to instantaneous pitch frequency. Using subjective listening tests it is shown that developed system provides high naturalness and intelligibility of reconstructed speech and can be applied to wideband and narrowband communication channels with various coding standards (including G.711 and GSM).