Onset and Frames & MT3 models: BPM?

[Joris Astier]

Hello,

We are using the Onset and Frames model and we would like to know how we could get the BPM of the converted audio in the output MIDI file?

Indeed, we just realized that the model did not provide BPM or PPQ in the MIDI file and all our MIDI editors displayed a BPM of 120 when we imported the MIDI file, which is the default BPM. But for many MIDI files generated via Onset and Frames, the original song is much faster than 120 BPM.

Would Onset and Frames or MT3 have a parameter to specify the BPM of the transcribed MIDI file?

If not, do you know of a tool that would allow us to do so?

Thanks in advance!

[Ian Simon]

Hi Joris, the Onset & Frames and MT3 models only transcribe note onset and offset timings; they don't attempt to infer higher-level properties like tempo, meter, chords, etc.

There are beat estimation models such as madmom (https://github.com/CPJKU/madmom) that operate directly on audio.  I've also noticed that if I load a MIDI file in MuseScore, it makes some attempt to automatically infer tempo and time signature; however, I have no idea how accurate it is.

-Ian

--
Magenta project: magenta.tensorflow.org
To post to this group, send email to magenta...@tensorflow.org
To unsubscribe from this group, send email to magenta-discu...@tensorflow.org
---
To unsubscribe from this group and stop receiving emails from it, send an email to magenta-discu...@tensorflow.org.

[Jay Peek]

Another way to get to the bottom of it is use corollary techs:

Using this premise of a rev eng Google Assistant designed recognize transients in human speech at hyper processing speeds: Artificial Digging: How Google’s AI Now Reveals What Producers Sampled (tracklib.com)

Sure it's outside of the Magenta environment but you need not look far to draw individual corollary tech to beat offset and onset in the tech you referenced.

I use email alerts for tech on the specific computer science I am in and then feed my head the proof-of-concept.

The brain works much faster if you simply look at tit on paper

POC X---> POC Y---> POC Z = close enough for gov't work.

Don't believe me

Chat GPT itself is based on the research axiom above (unofficially implemented by Elon Musk while with them)

https://binged.it/3LRohXt

GL ;-)

Jay Peek

@jpeek3456