This toolbox is a piano transcription inference package that can be easily installed. Users can transcribe their favorite piano recordings to MIDI files after installation. To see how the piano transcription system is trained, please visit: https://github.com/bytedance/piano_transcription.

Here is a demo of our piano transcription system: https://www.youtube.com/watch?v=5U-WL0QvKCg

The piano transcription system is developed with Python 3.7 and PyTorch 1.4.0 (Should work with other versions, but not fully tested). Install PyTorch following https://pytorch.org/. Users should have ffmpeg installed to transcribe mp3 files.

pip install piano_transcription_inference

Installation is finished!

python3 example.py --audio_path='resources/cut_liszt.mp3' --output_midi_path='cut_liszt.mid' --cuda

For example:

from piano_transcription_inference import PianoTranscription, sample_rate, load_audio

# Load audio
(audio, _) = load_audio(audio_path, sr=sample_rate, mono=True)

# Transcriptor
transcriptor = PianoTranscription(device='cuda')	# 'cuda' | 'cpu'

# Transcribe and write out to MIDI file
transcribed_dict = transcriptor.transcribe(audio, 'cut_liszt.mid')

Demo. Lang Lang: Franz Liszt - Love Dream (Liebestraum) [audio] [transcribed_midi]

If users met the problem of command line shows "Killed". This can be caused by users do not have sufficient memory.

We have built a large-scale classical piano MIDI dataset https://github.com/bytedance/GiantMIDI-Piano using our piano transcription system.

[1] High-resolution Piano Transcription with Pedals by Regressing Onsets and Offsets Times, [To appear], 2020