Spectral analysis of speech signals

摘要

The method of spectral analysis of signals, particularly speech signals, comprises the steps of deriving a time-varying first signal representative of the signal to be analyzed, filtering the first signal to obtain second time-varying signals each in one of a plurality of frequency bands, dividing the second time-varying signals into substantially equal units, dividing each of the second time-varying signals into a set of frames, each frame comprising at least one of said units. The position of the boundary between each frame is selected so as to minimise the "cost" of moving between frame boundaries. The subdivision of the signals should not blur or distort important acoustic or phonetic features of the signal and enables subsequent analysis of the signals to be performed in a way that is not computationally expensive. Application is to automatic speech recognition systems. IMAGE