Method of modification of the time scale of audio recordings, by playback at variable speed in real time

This paper presents a new method (SPEEDOVOX) which modifies the timescale of digitally recorded signals so as to obtain a speed-up ratio up to 4:1 on playback. The method preserves the formant structure, the perceived tonal quality and the original signal transitions, having the effect of providing better intelligibility and speaker recognition. It is based on the use of the signal energy and its variance to identify three typical regions: silence, transition and constant energy. The algorithm used for treatment of the regions is based on the SOLA method. Each phase of the method uses calculations implying exclusively the time domain, and their nature allows the process of continuous acceleration in real time to be sustained. The algorithm is optimised in order to reduce the execution time without compromising the signal quality in terms of intelligibility and speaker recognition. The results of application of the method on a 10-second recording are given