search

sony / hFT-Transformer

Pytorch implementation of automatic music transcription method that uses a two-level hierarchical frequency-time Transformer architecture (hFT-Transformer).
MIT License
73 stars 10 forks source link

request for colab inference notebook #2

Open GeorvityLabs opened 1 year ago

GeorvityLabs commented 1 year ago

@KeisukeToyama great work! hope you could create a google colab inference notebook, where we could use the colab t4 gpu. you can give the option for the user to load a .wav or .mp3 piano audio from their computer and then get the midi output using hFT-Transformer.

KeisukeToyama commented 1 year ago

Hello @GeorvityLabs Thank you very much for the request. I wish I could, but I'm afraid it is difficult to prepare a Google colab because I am busy with work. I appreciate your understanding.

GeorvityLabs commented 1 year ago

Hello @GeorvityLabs Thank you very much for the request. I wish I could, but I'm afraid it is difficult to prepare a Google colab because I am busy with work. I appreciate your understanding.

If you could give some instructions, I can try to make one

GeorvityLabs commented 1 year ago

Hello @GeorvityLabs Thank you very much for the request. I wish I could, but I'm afraid it is difficult to prepare a Google colab because I am busy with work. I appreciate your understanding.

If you could give some instructions, I can try to make one

Maybe once model checkpoint is released

KeisukeToyama commented 1 year ago

I'm wondering if you have read the README.md (https://github.com/sony/hFT-Transformer/blob/master/README.md)? The evaluation step explains how we can transcribe audio. We've already shared the checkpoint, so I think you can try it by yourself.

alex2awesome commented 10 months ago

there are a lot of steps in preprocessing audio files that are solely for evaluation (e.g. conv_note2label.py), not for taking a new audio file and generating MIDI.

It would be great if you could include a script that we could run starting with a new dataset.

Or, if it as simple as excluding 1 line of EXE-CORPUS-MAESTRO.sh, that would be great, too, if you could let us know. Just really hard to read through such a large code base and understand immediately what is going on, where

faltings commented 5 months ago

@GeorvityLabs Have you made one sample to load a .wav or .mp3 piano audio to get the midi output? Many thanks!