A variable frame-rate scheme for sinusoidal transform coding

Sinusoidal transform coding (STC) is known to be capable of producing good communication quality speech coded at bit-rates below 4kb/s. Discrete all-pole modelling (DAP) which can be more accurate than the conventional linear prediction (LP) analysis for voiced speech is adopted to improve the short-term spectral estimation with modification to accommodate the unvoiced speech in STC. A more robust frequency domain analysis-by-synthesis derived voicing cut-off frequency that divides the whole power spectrum into a lower voiced band and an upper unvoiced band enhances STC performance. In view of the different evolving characteristic of speech, in this paper, we propose a variable frame rate coding scheme by further investigating the potential reason of quality loss in reconstructed speech. This results in performance enhancement as well as bit-rate saving and leads to a more flexible and effective STC vocoder.