FBG Model Based Low Rate Coding of Speech

represented by a set of fbg filters each corresponding to a resonance peak where f is the formant frequency, b the 3 dB bandwidth and g the gain of a resonant peak. Speech coding algorithms are basic components in existing and future personal communication systems. Reducing the bit rate in transmitting speech is by way a matter of using slowly varying and numerical robust parameters to represent the speech. We are here focusing on parameters close to the physics of the speech production system for modelling the vocal tract. We have studied the dynamics (time variation) and numerical robustness of the new parameters (f formant frequency, b 3 dB bandwidth, g gain) and compared them with the traditional reflection coefficients. Listening tests indicate a possible reduction in bit rate of about two to three times for comparable sound quality. These fbgparameters will be coded, transmitted, interpolated and used as input into the receiver model in order to reconstruct the speech signal. Time(msec.) F re qu en cy (H z) red: 0, orange: −20, yellow: −40, green: −50