Master of Engineering
School of Electrical, Computer and Telecommunications Engineering
Gambino, Pasquale M.B., Pitch detection techniques for prototype waveform coding, Master of Engineering thesis, School of Electrical, Computer and Telecommunications Engineering, University of Wollongong, 1998. https://ro.uow.edu.au/theses/2407
Speech coding at the low bit-rate of 2400 bits per second requires an accurate and robust pitch period computed at low algorithmic delay to produce a perceptually satisfactory result. Low bit-rate speech coding algorithms such as those based on Prototype Waveforms can offer a speech quality comparable to higher bit rate counterparts. In this thesis two new algorithms, the 'Prototype Waveform Pitch Detector' (PWPD) and, the 'Dynamic Programming/Viterbi' (DPA^) Pitch Detector are introduced which can determine the pitch at frequent 5ms intervals. The PWPD is based on the 'Composite' Auto Correlation technique which sub-divides the speech frame into fixed sub-frames. The use of fixed sub-frames neglects to take full advantage of the variation in the pitch period. Within the paradigm of Prototype Waveforms, the PWPD achieves accurate pitch tracks by tracking the constituent autocorrelations across the speech frame, therefore maximising the detection of the pitch period and, at a significantly lower look-ahead delay, A further reduction in the look-ahead delay is achieved by the DPA^ Pitch Detector which is based on the 'Principle of Optimality' with a substantial extension of a variable state Viterbi-type Trellis. The inclusion of the trellis provides the capacity to maintain multiple candidate pitch tracks from which an optimal track (based on accumulated scores) is chosen.