First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Medium-Delay 8 Kbit/s Speech Coder Based on Conditional Pitch Prediction

Takehiro Moriya

NTT Human Interface Laboratories, Musashino-shi, Tokyo, Japan

A medium bit-rate (8 kbit/s), medium delay (10 msec one-way), and high-quality speech coder is designed. The coder uses a conditional pitch predictor in the framework of the backward adaptive CELP (Code Excited Linear Prediction) coder. This scheme transmits only 3 to 5 bits to select from the pitch period candidates pruned by backward pitch analysis. It also uses block-wise backward adaptive short-term LPC analysis and backward adaptive gain quantization. In coding experiments, Signal to Noise Ratio (SNR) and subjective quality were superior to schemes with conventional forward pitch prediction or without pitch prediction. Although the quality of the proposed coder is slightly inferior to that of the conventional forward CELP (more than 50 ms delay) at the same bit-rate, it can outperform conventional CELP if delayed decision of the excitation vector is introduced by paying a computational cost.

Full Paper

Bibliographic reference.  Moriya, Takehiro (1990): "Medium-delay 8 kbit/s speech coder based on conditional pitch prediction", In ICSLP-1990, 653-656.