Could it handle realtime transcription? about basic-pitch HOT 6 CLOSED

"Maybe" is a nice way of phrasing it :)

from basic-pitch.

Didn't bother since no info on whether it's even possible without costly retraining... I switched to this simpler method https://github.com/alesgenova/pitch-detection-app

from basic-pitch.

Good question. The inference in TF has been tested to be 3x faster than real time(I have to double check that number) if you're using the Typescript library (coming soon) on a mid-range chromebook.

However, the post processing step currently expects the output from fully processing the audio through the model. It may be possible to modify the post processing step to work on chunks at a time, but we haven't tested the approach and expect worse quality output if we were to try it.

from basic-pitch.

from basic-pitch.

I'm intending to have a crack at implementing this, as I need it for my own project (and can imagine it being useful in many applications). I understand there will be a loss in quality which for my application is not a problem since I don't need accurate note onsets or timing information.

Just wondering if there is a reason this is a "bad idea"? I don't think it will require retraining as I simply wish to extract note probabilities pre-post-processing.

Any insight appreciated!

from basic-pitch.

@supersational did you ever try this and could you share how it went, etc?

from basic-pitch.