Alonso Arevalo, Miguel A. (2006) Extraction d'information rythmique à partir d'enregistrements musicaux. PhD thesis traitement numérique du signal, ENST - TSI Traitement du Signal et des Images, ENST.

Full text available as: - these.pdf ( 3371 Kb )

Licence: Copyright Miguel A. ALONSO-AREVALO

Abstract

This dissertation presents work on the development of tools for the automated analysis of music signals, more exactly it proposes a number of techniques to identify and examine the fundamental elements of musical rhythm. There exist many applications requiring rhythmic information, for example automatic music transcription, music information retrieval, audio special effects, audio editing. The main research axis consisted in the development of techniques to estimate rhythmic parameters of music such as beat rates and beat localizations at two different metrical levels: the tatum and the tactus (or tempo). In this work, we have proposed to carry out the analysis after separating the audio signal in a deterministic part (containing the harmonic sounds) and the stochastic part (containing the residual signal after extracting the harmonic part from the original signal). We exploited the principle of the so-called Spectral Energy Flux, i.e., the rate of change of the power spectrum in audio signals as a function of time, to develop an effective algorithm to estimate the amount of musical stress at a given time. We have also presented a technique based on the dynamic programming algorithm which was specially conceived to track simultaneously the course of several rhythmic trajectories through time. The effectiveness of the whole system and of a number of its derivatives was validated using a database comprising 1435 musical pieces and covering ten musical genres. In addition, the developed algorithms were submitted to external evaluation within the framework of the international contest "Music Information Retrieval Evaluation eXchange (MIREX)" in the category of "tempo extraction", and in the edition of 2005 one of our algorithms obtained the first place among more than twelve submissions.

Table of content

1 Introduction
1.1 Musical rhythm
1.2 Metrical structure
1.3 Goals and dissertation outline
2 A survey on computational rhythm description
2.1 Computer rhythm analysis
2.2 General principle of computational rhythm description
2.3 Literature survey: current automatic rhythm analysis
2.4 Evaluation
2.5 Test corpus description
2.6 Conclusions
3 Estimating the degree of musical accentuation 55
3.1 Introduction
3.2 Pre-processing
3.3 Harmonic-plus-Noise decomposition
3.4 Calculation of the musical stress profile
3.5 Conclusions
4 Inducing rhythm metrics
4.1 Periodicity analysis
4.2 Pulse-period tracking
4.3 Beat location
4.5 Conclusion
5 System performance: results and discussion
5.1 On the effect of the window length and overlapping
5.2 Efficacy of the systemby musical genre
5.3 Influence of the H+N decomposition
5.4 Impact of the harmonic and noise parts in the accuracy
5.5 Influence of the frequency decomposition
5.6 Detection function comparison
5.7 Computational complexity
5.8 Beat-phase estimation
5.9 Tatum estimation
5.10 Conclusions
6 Concluding remarks and perspectives
6.1 Conclusions
6.2 Perspectives

Statistiques de consultation

Repository Staff Only: edit this item