A comparative study of various speech encoding schemes using vector quantization

Full search vector quantization (VQ) for dimensions up to 8 in the time domain and up to 12 in the frequency domain has been applied to various speech encoding schemes for real-time speech processing at rates between 4.8 kb/s and 9.6 kb/s. Our investigations include waveform coding (WVQ), adaptive linear predictive coding (APC), residual excited linear predictive coding (RELP), and the new concept of adaptive filtering of binary quantized speech (AFBQS) with telephone speech signals of male and female German and English speakers. The coding schemes were implemented on a HP-1000 computer combined with a FPS array processor AP-120B. The performance of the coding schemes has been evaluated on the basis of the SNR as well as by subjective listening tests. The results show that WVQ yields well intelligible and naturally sounding speech at low bit rates. For the other coding schemes VQ leads to additional bit rate reduction.