A Family of ADPCM Coders Implemented on Real-Time Hardware

Recently. work in ADPCM speech coding has focused on using the adaptive predictors, already included in the coder algorithm, as the basis for improving the perceived speech quality. The techniques use the CCITT adaptive predictors as the basis for noise shaping by error feedback in the transmitter and adaptive post-filtering in the receiver. The resulting signal sounds less noisy, although no actual improvement in signal-to-noise ratio is possible. The original work was performed with the intended application of improving the quality of 16 and 24 kbps ADPCM. In this paper we describe work we have done to extend this technique to lower bit rates, namely from 6 to 16 kbps. At the same time, we have implemented the algorithms on a single digital signal processor in order to evaluate their performance for a wide range of speech material. The bit rate is the product of the sampling rate and the quantization rate, the average number of bits used to encode a sample. In order to reduce the bit rate either the sampling rate or the quantization rate, or both, must be reduced. In order to reduce the sampling rate while maintaining compatibility with the standard 8 kHz sampling rate, we implemented digital interpolation. A sampling rate of 6 kHz offers a rate reduction of 25%. Standard telephone bandwidth of 200 to 3200 Hz needs to be reduced to 200 to 2800 Hz to implement a system of moderate computational complexity. Three different schemes were considered for reducing the quantization rate. A 25% rate reduction can be achieved by alternating one and two bit quantization. Another means of achieving the same reduction is to divide the difference signal into two bands and give the low band two bits while the high band gets one. A 50% quantization rate reduction can be achieved by using adaptive delta modulation.