Time scale modification of audio using Non-negative Matrix Factorization

Gerard Roma, Owen Green, Pierre Alexandre Tremblay
University of Huddersfield

Time scale modification is a widely used tool for music composition using existing recordings. The goal is often to modify the perceived timing of events while respecting other characteristics such as pitch, loudness and timbre. Given that sound events often overlap in real-world signals, source separation techniques are increasingly being investigated. In this article, we propose an algorithm for time scale modification of audio using non-negative matrix factorization, a popular tool for source separation. We demonstrate the use of this algorithm for preserving transients, as well as for changing the duration of an audio clip without modifying the envelopes of the sound events.

This page contains some audio examples of time-scale modification unsing NMF

Example 1: Glockenspiel (1.8 factor)

Example 2: Envelope preservation (0.6 factor)

Example 3: Novel artifacts (1.5 factor)

Example 4:Automatic rank (1.25 factor)