PURE SPEECH DETECTION USING VALLEY PERCENTAGE

摘要

A speech detection method detects pure-speech signal in an audio signal containing a mixture of pure-speech and non- or mixed-speech signals. The method detects the pure-speech signals by computing a novel Valley Percentage feature, a measurement of the low energy parts of the signal, and performing a threshold decision on this feature. The method further employs a morphological closing filter to eliminate unwanted noise prior detection, and after, a combination of morphological closing and opening filters to remove aberrant pure- or non-speech classifications resulting from impulsive audio signals, in order to more accurately detect the boundaries between the pure- and non-speech portions of the signal.