Abstract:

To improve efficiency of the waveform similarity overlap-and-add (WSOLA) algorithm in audio signal processing at high sampling rate, this paper proposes a matching method that is progressively performed from the short time mean envelop to the signal waveform. We compute a rough matching envelop based on short time mean envelop, and then perform an exact waveform matching for time-scale modification of the audio signal. The
algorithm reduces computation complexity, and improves efficiency with good outcome. In addition, the length of
analysis windows is dynamically adjusted based on the modified low energy ratio parameter. Experiments show that it significantly improves processing results of mixed audio.

Key words: time-scale modification of audio signal, time-scaling factor, short time mean envelop, cross-correlation coefficient, modified low energy ratio