Piano transcription inference

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.

Demos

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

Installation

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!

Usage

Want to try it out but don't want to install anything? We have set up a Google Colab.

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

This will download the pretrained model from https://zenodo.org/record/4034264.

Users could also execute the inference code line by line:

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', checkpoint_path=None)  # device: 'cuda' | 'cpu'

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

Visualization of piano transcription

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

FAQs

This repo support Linux and Mac. Windows has not been tested.

If users met "audio.exceptions.NoBackendError", then check if ffmpeg is installed.

If users met the problem of "Killed". This is caused by there are not sufficient memory.

Applications

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

Cite

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