stengerh
Incorporate new study for finding beets
Hi,
Thanks for this fine piece of software, detecting slow music and classical music still seems to be off a lot. Probably it's my fault as I can't seem to graph the parameters and how to use them correctly.
In looking for other BPM I found this https://www.researchgate.net/publication/3334132_A_Tutorial_on_Onset_Detection_in_Music_Signals which maybe can make this BPM better.
Or if it's not the case, maybe an explanation of which values to use for slow, classical pop music to get better results.
Hi, I am not the original author of the component. Essentially I only applied some polish to the component like technical updates, performance tweaks, and resolving license conflicts. Unfortunately I lack the musical training to do BPM labeling by hand. Therefore I don't feel qualified to provide hints on this topic. I suggest you ask your question on the foobar2000 forum. You can find the link in the readme file or in the about message of the component. You will have a much better chance of getting helpful advice from fellow users there. :)
While I cannot comment on the used algorithm without checking the code, I noticed that your paper is referenced in Onset Detection Revisited from 2006 which was one of the references used by the original other of the component. You might be interested in the conclusion in that paper and the description of the data sets used for evaluation. In particular the authors of Onset Detection Revisited observe that there is no single best algorithm among the ones they tested. Some work best on popular and jazz music, others on solo violin or piano music. I could not find any mention of classical music played by an orchestra.
If you check citations of your paper you can find some more recent ones, for example
- Chowdhury, Shreyan & Guha, Tanaya & M. Hegde, Rajesh. (2017). Music Tempo Estimation Using Sub-Band Synchrony. 3093-3096. 10.21437/Interspeech.2017-1000. (PDF)
- Percival, Graham & Tzanetakis, George. (2014). Streamlined Tempo Estimation Based on Autocorrelation and Cross-correlation With Pulses. Audio, Speech, and Language Processing, IEEE/ACM Transactions on. 22. 1765-1776. 10.1109/TASLP.2014.2348916. Source code for the used tools are available at https://github.com/marsyas/marsyas . Compiled binaries seem to be available from the MARSYS homepage. This might be great for experimenting with your own audio files although the license would not allow an inclusion of their C++ library into the component.
