piano-transcribe

Original code is provided by the team from ByteDance [pdf, code]

Usage

1. Install dependencies:

pip install -r requirements.txt

• Different versions of packages may result in errors when running.

2. You may want to change the designated GPU for transcription in inference.py

3. If you have multiple GPUs to run, you can either:

   • Uncomment lines 58-62 in piano_transcription_inference/inference.py to run parallel on multiple GPUs for faster transcription
   • Run with the original package directly:

   pip install piano_transcription_inference
   

4. To start transcribing with pre-trained model:

   • Set the audio_path to the directory of audio files, the transcribed result will then all be saved to a new directory midi

   python transcribe.py --audio_path <audio_folder> --output_midi_path ./midi
   

   • If running into the error of audioread.exceptions.NoBackendError, install ffmpeg with conda:

   conda install ffmpeg