Speech formant trajectory pattern recognition using multiple-order pole-focused LPC analysis

A technique termed pole focusing is presented that provides a novel approach to obtained high-resolution formant data for pattern recognition processing of the short-time speech spectrum. The technique offers reliable detection of weak nasal formants and formants undergoing rapid transitions in frequency, areas where parametric spectral analysis typically performs poorly. Much of the recognition of speech using a feature-based approach relies heavily on the detection of formant time-frequency trajectory patterns, which gives the identification not only for the voiced speech sound currently under analysis, but also can provide important cues to pre- and postvocalic speech. The enhanced formant detection properties offered by pole focusing therefore can considerably improve the reliability of formant pattern recognition