New methods for improvement of sinusoidal transform vocoders
F.A.R. Nascimento, Francisco J. Fraga · 2005
New methods for improving the speech quality of sinusoidal transform vocoders at 2.4 kbps are presented. Pitch estimation refinements are implemented, since its accurate estimation is essential for the efficiency of the harmonic sine-wave speech model. A more efficient procedure for estimation of the spectral envelope parameters, which exploits the subjective loudness psychoacoustic characteristics of the human hearing system and an alternative approach for the synthetic phase composition are proposed, so that a better computational cost is obtained. Experimental results indicate that the use of all these methods enhances the performance of a sinusoidal transform vocoder, operating at 2.4 kbps, improving its subjective quality and computational cost.