We have improved G.728 output speech quality for frame erasure channels. Three cases are considered: (1) no change to G.728, (2) change only the G.728 decoder, and (3) change both the encoder and decoder. In case 1, we synthesize a bit-stream during erased frames so that the decoder decodes an excitation with low energy or with characteristics similar to the excitation of previous good frames. In case 2, the gain-scaled excitation and LPC coefficients are extrapolated, and vital operations of backward LPC and gain adaptations are continued. Case 3 adds spectral smoothing and increases bandwidth expansion for the LPC and gain predictors. These techniques are quite effective, as the speech quality degradation due to 1% frame erasures ranges from just slightly noticeable in case 1 to almost unnoticeable in case 3. For case 3, the output speech is still intelligible for frame erasure rates up to 10% or even 20%.

Adaptive quantization is used in many ADPCM systems to improve performance. The authors present an analysis dealing with the effect of adaptive quantizers on the stability of ADPCM systems as compared with ADPCM with nonadaptive quantization. Stability of the system is noted to equate with recovery from encoder internal state discrepancies. A sufficient condition for stability that leads to a relation between adaptation rate and stability is presented. This indicates that a price is paid for adaptation in terms of more stringent restrictions on the predictor to guarantee stability. The theorem further indicates that stability is related to the maximum rate of quantizer step size decrease and can be viewed as a theoretical justification for the shape of the Jayant "one-word memory" multiplier curve for adaptive quantization.>

Discusses the use of a variable bit rate ADPCM system for speech coding. The authors present an ADPCM system, using arithmetic coding and Kalman filtering techniques, that can code signals at very low bit rates, where standard fixed bit rate ADPCM systems fail. The proposed system, at an average bit rate of 16 kbps, produces a reconstructed speech signal that by informal listening tests has been judged to be of significantly higher quality than the LD-CELP 16 kbps speech coding standard. The authors believe that the output quality is comparable to that of LD-CELP at an average rate of around 12 kbps, and the quality at 8 kbps is very promising. Other advantages of the scheme are a reduction in complexity over the LD-CELP standard, and an increase in design flexibility.>