Pitch-Scale Modification Based on Formant Extraction from Resampled Speech

 Abstract

Pitch-scale modification that can change the tone and the prosody of speech is useful in privacy protection and entertainment. One of the approaches for pitch-scale modification is the analysis-synthesis method. It has the freedom for synthesizing arbitrary voice once the speech parameters such as LPC coefficients and residual signal are obtained.

In this paper we propose a pitch-scale modification method based on formant extraction from resampled speech. The formant, which is the spectrum envelope of speech signal, can be extracted by LPC analysis, and this procedure, so-called de-formant, eliminates the short-term correlation incurred by vocal tract filter. The frequency response of LPC synthesis filter determines the timbre of synthesized speech. The residual signal mainly consists of long-term components, the pitch harmonic, which determines the tone of speech and can be easily modified by using the resampling technique. A dual-resampling mechanism is used to obtain the modified formant and modified pitch harmonic, respectively. The pitch-scale modification mentioned above is only performed in voiced frames because they have high energy and are relatively stable.  And the cross-correlation coefficients are calculated to locate the synchronization point, i.e., the pitch mark. Experimental results show that the speech can be successfully modified to different timbre and tone with high quality.