A Very low bit rate Phonetic Vocoder for Farsi Language:

Abstract This paper presents the research steps for design and implementation of a 300 bits/s very low bit rate phonetic vocoder. In coder side of this vocoder, recognized phonemes, phoneme duration, voiced-unvoiced, pitch frequency and gain are determined and after quantization are sent to decoder. Phoneme recognizer is based on Hidden Markov Model and is trained using Farsdat database. In this research many experiments are conducted to improve the performance of vocoder in phoneme recognition, voiced-unvoiced estimation, pitch and gain detetection, quantization of vocoder parameters and specch synthesis. Improvements in phoneme recognition led to performance improvement from 57.6% to 77.0%. Result of MOS tests on comprehensibility was 2.7.