02-03-2013, 12:10 PM
New speech harmonic structure measure and its applications to speech processing.
Abstract
The harmonic structure can be easily recognized in time-frequency representation of speech signals in adverse environment. The harmonicity is a measure of the completeness of a harmonic structure. This paper presents a new harmonic structure measure that extends the conventional harmonicity to a set of harmonicities. They are expressed in terms of the grid harmonicity, the temporal harmonicity, the segment-spectral harmonicity, and the segmental harmonicity. The grid harmonicity measures the completeness of individual harmonics in each frame. The grid harmonicities in a frame are summed up to form a temporal harmonicity for representing the strength of harmonicity. The segment-spectral harmonicity, computed by summing specific grid harmonicity over a segment, evaluates the integrity of individual harmonics across a segment. The segmental harmonicity evaluates the total strength of harmonic structure within a segment. This set of harmonicities is available for a systematic analysis of the harmonic structure and effective to several speech processing tasks. The applications to speech distortion analysis, robust fundamental frequency estimation, robust voicing detection, and speech enhancement are demonstrated.