Design and development of a very low bit rate Phonetic vocoder for Farsi Language

This paper presents a very low bit rate (VLBR) speech vocoder for Farsi language. Vocoder encoder consists of a phoneme recognizer, voiced/unvoiced, pitch and gain estimators. Phoneme index, state durations, pitch and gain are quantized, encoded and transmitted to decoder. In the decoder, phoneme HMMs are concatenated according to the phoneme indexes, and a sequence of MFCC coefficient vectors is generated from the concatenated HMMs by using an ML-based speech parameter generation technique. Finally we obtain synthetic speech by exciting the MLSA (mel log spectrum approximation) filter, whose coefficients are given by MFCC coefficients, according to the pitch information. A phoneme recognition rate of 77% and a total bit rate of 295 bits/s were obtained. Intelligibility and naturalness of synthesized speech was evaluated using MOS subjective tests. MOS intelligibility score was 2.7