Low rate speech representation by vector quantizing transform components

A transform-based system is described for vector quantizing speech at 4800 BPS. Multiple codebooks were used to quantize groups of harmonic components of speech, corresponding to different frequency bands. To ensure that the distortion caused by the vector quantizer at low frequencies was kept at a minimum, low-dimensionality codebooks were used. Furthermore, to reduce the errors of the FFT resolution, transform components, within the baseband and adjacent to the pitch-harmonic components, were also encoded and quantized by low-dimensionality codebooks. This was found to improve the quality of speech. Speech produced at 4800 BPS using this method was found to be of good quality, particularly for female speakers. This is because the number of harmonic components required for female (high-pitch) speakers is generally smaller than that required for male (low-pitch) speakers.