search

michaelkrzyzaniak / Beat-and-Tempo-Tracking

Audio Beat and Tempo Tracking
MIT License
94 stars 12 forks source link

Tempo tracking issues for jittery, feel-based human tempos #12

Open pomplesiegel opened 10 months ago

pomplesiegel commented 10 months ago

Hello! Great library - I'm having a lot of fun with it.

One area I am struggling with to find the right parameters, pre-processing, wrapping around in order to have good performance is for songs with a real human-based swagger going on. On these I'm finding the algorithm confidence hovers around 20-25%, in contrast to other other songs wherein it performs well.

A good example of this is George Clinton's Lickity Split. As a human it's easy for me to tap out this tempo, and it feels quite solid, hovering somewhere around 110-112 bpm the whole song. However, in my testing the algorithm is not very confident what's going on. It sometimes finds the correct tempo, but often detects 143 or 156 bpm, and always with fairly minimal confidence.

I'm interested in your experience running the algorithm against this song. I'm using a microphone to detect the tempo being played by speakers, but perhaps it would be similar with the actual sound file.

Are there any parameter changes you would recommend / are willing to test that allow a bit more "easygoing" of an attitude in regard to beat placement, so human players have a bit more leeway in the algorithm's eyes?

Looking forward to your thoughts!

Thank you, Michael

pomplesiegel commented 8 months ago

Hello @michaelkrzyzaniak, I hope you're doing well! Any thoughts on the above? I'm trying to get it to behave a bit better for these types of music but hitting a bit of a wall. Curious if you have some insights / suggestions. Thank you!

pomplesiegel commented 6 months ago

Hi @michaelkrzyzaniak , hope all is well with you! Trying one more time, just to see if you have any thoughts on the above. Thanks!

acrilique commented 5 months ago

Hi, haven't played a lot with the library myself... However, playing that song samples' through the library myself with a gaussian histogram decay of 0.9999 (higher than default) I get an pretty solid tempo of 100-107bpm. Sure, it is not like the results you get when you play straight-tempo electronic music but seems pretty good of a result to me. What parameters have you tried?

image

pomplesiegel commented 5 months ago

Thanks for trying this out! The key issue I'm seeing is that the actual tempo is 111-114 throughout the song (you can check it by using the LiveBPM app or tapping with a metronome app). The only way I was able to get it it to produce the right answer was to have it lock onto half of that tempo (55-56 bpm) and double the result. However it's reluctant to go that low.

acrilique commented 5 months ago

Out of curiosity and in order to get the best parameters for different kinds of music, I'm writing a small program to test all of this live, with different audio files. Just finished the base program. I will add options to change all parameters tomorrow. Hope you find it useful or at least fun. I'm open to suggestions and issues/pull requests. https://github.com/acrilique/TempoTest

pomplesiegel commented 5 months ago

That's amazing! Thank you so much for your work. Looking forward to trying this out once you think it's ready for testing. Best regards, Michael

On Sat, May 25, 2024 at 7:38 PM Lluc Simó @.***> wrote:

Out of curiosity and in order to get the best parameters for different kinds of music, I'm writing a small program to test all of this live, with different audio files. Just finished the base program. I will add options to change all parameters tomorrow. Hope you find it useful or at least fun. I'm open to suggestions and issues/pull requests. https://github.com/acrilique/TempoTest

— Reply to this email directly, view it on GitHub https://github.com/michaelkrzyzaniak/Beat-and-Tempo-Tracking/issues/12#issuecomment-2131688558, or unsubscribe https://github.com/notifications/unsubscribe-auth/AALCC4CIVR65EYI5JEXK56LZEEOHVAVCNFSM6AAAAABBYRQFHSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMZRGY4DQNJVHA . You are receiving this because you authored the thread.Message ID: @.*** com>

acrilique commented 3 months ago

Hi @pomplesiegel , after a migration to gtk4, the tester is ready for use. Next functionality will be a way to plot the detected tempo vs time and to save the results to a document, along with the values of the parameters in time. However, I think that I'm going to migrate away from gtk4 because I want the app to be cross-platform and I'm not able to run it on windows (did build it, but I'm getting gdk errors at runtime, gdk being the graphics backend for gtk). Anyway, If you're on Linux or MacOS, please try to build and run the app in its current state.