A new lpc vocoder model for low bit rate speech coding

The problem of representing digital speech signals using as few bits as possible is becoming increasingly important in the modern era of digital communications. One particularly efficient speech coding algorithm called the LPC vocoder uses a fully parametric model to mimic human speech. Since periodic update of the model parameters requires fewer bits than direct representation of the speech signal, this vocoder can operate at very low bit rates. Unfortunately, the speech output from an LPC vocoder is not acceptable for many applications because it does not always sound like natural human speech, especially in the presence of acoustic background noise. This thesis develops a new, fully parametric, LPC vocoder model for efficient digital storage and transmission of speech. Unlike previous speech models, the new model has a flexible parameterization of the speech excitation signal and can produce natural sounding synthesized speech, even in the presence of acoustic background noise. A speech coder based on this model has been implemented in a real-time system for further experimentation and performance evaluation. Formal listening tests have shown that this new 2400 bps LPC vocoder performs better than the current state of the art for both clean and noisy input speech. In addition, development of the model has led to some insights into human perception of speech signals. This increased understanding may also be useful in other areas of speech processing.