Assignment 3 - Tempo Induction and Beat Tracking for Audio Signals


Annotated Bibliography

Allen, P., and R. Dannenberg. 1990. Tracking Musical Beats in Real Time. In Proceedings of the 1990 International Computer Music Conference, International Computer Music Association, 140-3.
This paper describes the implementation of a system that described the concept of a beat having a period and phase, and introduced the 'beam search' idea, where multiple beat theories are conducted by a set of agents, which are adjusted, created and deleted. It is not clear whether the system was implemented with MIDI or audio as input.

Davies, M., and M. Plumbley. 2004. Causal Tempo Tracking of Audio. In Proceedings of the 5th International Conference on Music Information Retrieval (ISMIR 2004), 164-9.
This paper presents a reltime audio beat tracking system that has been implemented to perform with different musical styles and be robust in terms of accepting tempo changes in the musical piece. Whereas tempo induction and beat tracking techniques are similar to those implemented in previous systems, the paper implements different techniques in note onset detection at preprocessing stage (a combination of high-frequency analysis and complex spectrum analysis to accomodate sounds both with and without percussive content).

Dixon, S. 2001. Automatic extraction of tempo and beat from expressive performances. Journal of New Music Research, 30, (1): 39-58.
This paper describes a tempo induction and beat tracking system (with a clear division between those two stages of tempo analysis) that accepted both MIDI and audio as its input. Note onsets are used for tempo induction, with a clustering algorithm determining significant clusters of data. After a ranked list of potential tempo values has been formed, the beat tracking process implements multiple hypothesis approach to determine the beat phase. Musical salience, an example of higher-level musical knowledge rule-based approach, is used to determine the most likely candidate for beat phase.

Goto, M. 2001. An Audio-based Real-time Beat Tracking System for Music With or Without Drum-sounds. Journal of New Music Research, 30, (2): 159-71.
This article presents a realtime beat tracking system, which was implemented as a combination of techniques used in the previous two systems developed by the author (for piece with and without percussive sounds), making it more generalized in terms of the audio input. The system is distinguished through extensive use of higher-level musical knowledge-based approach, with musical knowledge, drum pattern knowledge and chord structure knowledge being the three main rule categories. The system is limited to the western music style, but very effective for it.

Goto, M., and Y. Muraoka. 1995. A Real-time Beat Tracking System for Audio Signals. In Proceedings of the 1995 International Computer Music Conference, International Computer Music Association, 171-4.
This article describes a system (which is one of the two original systems designed by the authors) that deals with beat tracking for sounds with strong percussive content. The drum patterns are extracted from the audio signal and matched against a set of predetermined drum pattern templates. This system was later combined with the non-percussive sound analysis system into a combined general beat-tracking implementation.

Scheirer, E. 1998. Tempo and beat analysis of acoustic musical signals. Journal of the Acoustical Society of America, 103, (1): 588-601.
This article deals with a beat-tracking system that was based on tuned resonators. The signal was split into six frequency bands, and amplitude envelopes that were extracted from the bands were passed through a set of 150 comb filters, with each filter representing an individual possible tempo value on a discretized scale. The outputs of the filters were then summed across the frequency bands, and the highest output value determined the tempo and the phase of the beat. One of the issues that were present in the system was the need to fine-tune the filter spacing in order to produce a high quality result in terms of tempo representations.

Schloss, W. 1985. On the Automatic Transcription of Percussive Music: From Acoustic Signal to High Level Analysis. PhD thesis, Stanford University, CCRMA.
This is one of the earliest works on beat tracking of audio signals. The system used the technique of high-frequency content analysis for onset detection at the preprocessing stage -- it detected note onsets as peaks in the slope of a high-pass filtered signal.