Matthew Davies: "Automatic Rhythm Analysis of Musical Audio: Techniques and Applications"

Matthew Davies is a Postdoctoral Researcher, Queen Mary Electronic Engineering and Computer Science (London).

In this seminar I will discuss the automatic extraction of rhythmic and metrical information from audio signals. This will cover three main tasks: the extraction of beat times, finding measure boundaries and the derivation of a predominant rhythmic pattern to summarise rhythmic structure.

Beat times are extracted from an onset detection function through two processes. The first uses Viterbi decoding to track the tempo across the input. This is followed by a dynamic programming algorithm to align beats to peaks in the onset detection function.
The measure boundaries are recovered by finding the spectral difference between beat-synchronous analysis frames, and the predominant rhythmic pattern by clustering measure length onset detection function frames.

To illustrate the combination of all three layers of analysis I will present a rhythm morphing technique which can transform the rhythmic structure of one musical input to take on that of another.