A Comparison of thne Performance of Four Low-Bit-Rate Speech Waveform Coders

The quality of the reproduced speech from waveform coders can usually be improved by increasing their complexity. However, increasing the complexity also usually increases the cost. The practical problem in choosing a coder, at the current state of the art, becomes one of knowing how much loss in quality is sacrificed when opting for a less complex (less expensive) coder. This paper compares subjective quality ratings for four different speech waveform coder algorithms of varying complexity. The algorithms, rated in order of their complexity are: A D P C M - F (adaptive differential pulse code modulation with a fixed predictor), SBC (subband coding), ADPCM-v (adaptive differential PCM with a variable predictor), and ATC (adaptive transform coding). Each of these four algorithms was studied at three different transmission rates: 24, 16, and 9.6 kb/s. These coders were chosen because they represent a number of different classes of coding techniques ranging from relatively simple (inexpensive) to highly complex (costly) schemes. The choice 699