Linear prediction of formants for low bit rate digital speech transmission

A new method based on linear prediction technique in transform domain is proposed in this paper to achieve low bit rate digital speech transmission. Seperate processing of voiced/unvoiced sections of speech is employed. In the case of voiced sounds the first formant is chosen from the short time sequency spectrum (1) and then a prediction coefficient is calculated using LP techniques from which the other two formants can be predicted at the receiver end. In the case of unvoiced sounds, components from the mid sequency spectrum region are chosen and transmitted. Computer simulation of the system using Walsh-Hadamard transform results in intelligible speech recovery at 1.2 Kb/s. The proposed system is simple to implement compared with existing systems of similar transmission rates.