Subjective and Objective Evaluation of Pitch Extractors for LPC and Harmonic Deviations Vocoders.
Vijay Viswanathan, William Howard Russell · Defense Technical Information Center (DTIC) · 1984
This report describes the work performed on subjective and objective evaluation of pitch extractors for use in 2.4 kbit/s LPC and harmonic deviations (HDV) speech coders. From a review of the available pitch extractors, five algorithms were chosen and implemented. An existing algorithm was modified to extract automatically accurate, reference pitch and voicing data from the subglottal signal recorded simultaneously with the speech signal using a miniature accelerometer; this data was used for objective evaluation of pitch extractors. Since the accelerometer is relatively insensitive to acoustic background noise, this method yields accurate pitch and voicing data even in noise. For formal subjective evaluation of the chosen pitch extractors, a speech database of 48 sentences that are likely to cause pitch and voicing errors was developed. The objective evaluation method developed in this work involves comparing, on a frame-by-frame basis, the test pitch extractor data with the reference pitch data, computing objective pitch and voicing error measures, and averaging over the sentences from the speech database. (Author)