Trellis code excited linear prediction (TCELP) speech coding
Cheng-Chieh Lee, Yair Shoham · 2003
This paper describes using the trellis-based scalar-vector quantizer for sources with memory to solve the excitation codebook search problem of code excited linear prediction (CELP) speech coders. This approach leads to a 24 kbit/s telephony-bandwidth low-delay (3 msec) trellis CELP coder, which outperforms both ITU-T 15 kbit/s G.728 LD-CELP and G.726 32 kbit/s ADPCM. Since the codebook is derived from a scalar alphabet, the proposed coder can effectively handle excitation vectors in the 24-dimensional space (to realize considerable vector quantization gains) and has a computational complexity of approximately 75% of that of ITU-T G.728 LD-CELP.