Efficient speech coding techniques by simplified orthogonal transformation
Jing‐Chung Juang, Lin‐Shan Lee · Journal of the Chinese Institute of Engineers · 1985
Digitally represented speech has to be quantized into as small number of bits as possible to save the storage memory or transmission bandwidth. Such lower bit rates can be achieved by encoding the parameters characterizing the special features of speech singals instead of trying to encode the speech waveform directly. Linear predictive coding (LPC) is one such technique. Selective coding is to allocate different numbers of bits to different parameters with different statistical variance. Orthogonal transformation can be applied to LPC parameters here to more efficiently allocate the bits. Because orthogonal transformation requires high complexity of computation, a “simplified orthogonal transformation” is proposed in this paper, which can be obtained by a series of training data, to tremendously reduce the computation but essentially preserve the advantages. Many efficient coding techniques in such case are proposed in this paper, the results indicate that very satisfactory speech quality can be achieved at bit rate as low as 1,200 bps (if only scalar quantization can be used) or 750 bps (if vector quantization can be used). Computer simulation strongly supports the potential of these techniques.