An adaptive transform coder is proposed that makes effective use of vector quantization. It is aimed at medium-band (4. 8-9. 6 kbps) speech coding. In order to minimize waveform distortion, DFT (discrete Fourier transform) components of the LPC (linear predictive coding) residue are quantized using two novel techniques: bit allocation to blockwise DFT components has been made adaptible to obtain snr gain, and quantization has been divided into two stages (vector-scalar) for higher bit rates. Subjective quality of speech coded at 7. 2 kbps is comparable to that for 5 bit log-PCM. Intelligibility tests show that the coder is resistant to background noise.
|Number of pages||11|
|Journal||Denki Tsushin Kenkyujo kenkyu jitsuyoka hokoku|
|Publication status||Published - 1985 Jan 1|
ASJC Scopus subject areas