Recently, variable-rate speech vocoders have been studied in various aspect to achieve high synthetic speech quality. As the need to transmit various type of data via the same channel arises, some problems must be solved because each channel has its own bandwidth. The method that assigns fixed transmission rate to each data is practical but wastes the performance of transmission capability.
This paper proposes MMP-MLQ (Modified Multi Pulse-Maximum Likelihood Quatization) vocoder, which produces high synthetic speech quality with variable rate below 6.3 kbit/s. Transmission rate is determined by the number of pulses. This vocoder uses a modified linear prediction method to find the coefficient of speech synthesis filter. The method used in this vocoder represents spectral robustness against the position of analysis frame of input speech data.
The quality of synthetic speech produced by the proposed vocoder was compared with that of the dual rate speech coder for multimedia communication. The proposed one manifests better synthetic speech quality than that of the dual rate speech coder measured in CD (Cepstral Distance). In particular, the proposed vocoder can be operated in more possible transmission rate.