bambooonfire/piano-ai-transcription

Some interesting transcription model wrapper

Piano-AI-Transcription

This is a simple Python "wrapper" that utilizes ByteDance's amazing polyphonic transcription tool. The model is not widely-known but it definitely is a powerful tool. In my opinion, no other models beat this. It supports 1) Correct timing, 2) Correct MIDI velocity, 3) Partially correct sustain points.

It also supports direct YouTube URL input so you don't have to use shady sites to download audio.

Source Model

ByteDance: https://github.com/bytedance/piano_transcription

Usage

1. Download this repository.
2. Install python3.7 from https://www.python.org/downloads/
3. Drag any audio file to the Input folder.
4. Or you can follow the instructions in command line and directly render Youtube URLs. (Pretty handy right?)
5. Open terminal and type python3.7 then press SPACE
6. Drag RUN.py and press ENTER
7. MIDI will be exported to the Output folder.
8. Currently, the python script will automatically remove the audio files in Input to save storage.

Examples

The Input and Output folder in this repo already contains an example render of Charles Cornell's sexy "Heart and Soul" jazz piano performance.

The output file is in MIDI format and you can use any DAW and VST to render the audio yourself.

piano-ai-transcription's Issues

Error message

What can I do about this error?

TypeError: guvectorize() missing 1 required positional argument: 'signature'