Home Papers Reports Projects Code Fragments Dissertations Presentations Posters Proposals Lectures given Course notes

How to measure Tempo ? Autodifference, Autocorrelation, Rhythm Extraction and Neural Networks

Werner Van Belle1 - werner@yellowcouch.org, werner.van.belle@gmail.com

1- Yellowcouch;

Abstract :  The peaks/valleys in autodifference/autocorrelation plots correspond to the potentially perceived tempo of music. Consequently, they can be used to measure the tempo of music. The problem with these techniques is a certain bias, which is unique for each song and without its removal, reported tempos might be off by a multiplication factor. In this presentation we discuss removal of such biases. Secondly, we discuss how the problem of tempo-harmonics (120 BPM reported as 90 BPM for instance) can be reduced from a multi-class classification problem to a binary classification problem. We also argue that the use of 'beats per minutes' is an artificial measure and should be superseded with the use of 'measures per minute', as it turns out to be a much more robust measure. In the second part of the talk, we will focus on a variety of techniques to extract rhythmical information and classify time-signatures. We do so because the combination of time-signature and tempo forms the key to solve the tempo-multiplication problem. Although from a meta-scientific point of view, we know that a neural network should be able to classify various time-signatures, the implementation of such network is still somewhat elusive because it is unclear how to present the data to the network. We discuss a variant of bayesian classification as well as a number of neural networks we tried. Lastsly, we discuss the potential problems of creating such classifiers if we take into account the potential amount of data one might need to train a neural network properly.

Keywords:  tempo extraction, bpm measurement, bpm counter, rhythm extraction, rhythm pattern extraction, rhythm classification
Reference:  Werner Van Belle; How to measure Tempo ? Autodifference, Autocorrelation, Rhythm Extraction and Neural Networks; Presented at Technische Universit├Ąt Dortmund; Germany; October 2010
See also:
Presentation given at the Chaos Communication Camp
A paper on the rhythmic pattern extraction
BpmDj homepage
A number of strategies to speed up tempo measurement
Biasremoval from BPM Counters and an argument for the use of Measures per Minute instead of Beats per Minute [article | slides]
Basic Autodifferencing Explained