ISCA Archive Eurospeech 1995
ISCA Archive Eurospeech 1995

Efficient multiband excitation linear predictive coding of speech at 1.6 kbps

W. M. E. Yu, Cheung-Fat Chan

A 1.6 kbps MBE speech coder is proposed in this paper. Several improvements on the conventional MBE coder have been made in order to maintain the quality and intelligibility of the original speech at 1.6 kbps. We developed an improved pitch detection algorithm which results in more accurate pitch estimation for natural speech synthesis. The number of bits for coding V/UV information is significantly reduced by employing a simplified V/UV mixture function which requires coding of the position of a V/UV transition frequency only. Band magnitudes are coded via linear predictive model. An efficient recursive procedure was developed for computing and quantizing the 2DdLSP residuals simultaneously. Subjective speech quality was further improved by the application of a postfilter in the frequency domain. The speech coder was implemented in real-time on the TMS320C30 DSP chip.