Pitch determination of speech signals in the presence of noise

The research described in this thesis has been carried out in three specific areas, namely the implementation and modification of pitch determination algorithms for speech signals, an investigation of the performance of the algorithms under different conditions of acoustic noise, and the real-time implementation of one of the algorithms using a high speed single-chip digital signal processor. The thesis presents a discussion on the applications of pitch determination algorithms for speech, followed by a description of the mechanism for human production and perception of speech signals. A modelling technique for the speech production mechanism is also introduced. Pitch determination algorithms for speech signals are discussed in detail, with particular reference to the two algorithms implemented during this research. Aspects of their implementation on a digital signal processing system based on an IBM PC-XT and commercially available signal processing software are described. Several important modifications to enhance the algorithms' performance have been carried out, and these are discussed in the thesis. Extensive tests under varying conditions of acoustic noise have been performed, and these are presented in the thesis. Results are presented and discussed, giving an indication of the effect of the noise on the performance of the algorithms. Finally, an implementation of a time domain pitch determination algorithm using a DSP device is presented, with detailed notes on the various aspects of the implementation. Results indicate that using appropriate AID hardware, the algorithm can achieve realtime pitch determination of speech