Digital Coding of Speech with Zero Delay

Abstract

Algorithms for low bit rate voice typically depend on the use of encoding delay to permit efficient designs of quantizes, predictors and bit allocation techniques. The lowest bit rate at which high quality coding with zero delay is generally recognized as possible is 32 kb/s, the bit rate of standard ADPCM. This paper describes a technique for enhancing ADPCM quality, based on backward-adaptive zero-delay algorithms for noise shaping and post-filtering. At 16 kb/s, this technique provides very good communications quality, with a mean opinion score approaching 3.5 on a scale of 1 to 5. This is less than half a point away from the measured performance of much more sophisticated algorithms at 16 kb/s, and is therefore a very strong result for a zero-delay coding algorithm. While an encoding delay of zero may not be a critical need for most applications, the successes of our noise-shaping algorithms suggest new possibilities for low delay coding of speech with high quality. They also suggest ways of enhancing the quality of wide classes of high-delay algorithms in the bit rate range of 4 to 16 kb/s.

Indexing terms:

• Coding of speech
• Encoding delay
• ADPCM Coder