MIDIHarm - a PSOLA-based MIDI vocal harmonizer

Python tool for pitch-shifting MP3 files to MIDI files.

Usage

1. Clone the GitHub repo to your computer with git clone https://github.com/sonofthomp/midi-harmonizer.git. Run cd midi-harmonizer to enter the directory.
2. Install the package requirements with pip install -r requirements.txt
3. The harmonization happens in two steps. First, generate a "tone file" -- this will be an MP3 file which you should use a guide to sing the melody of the song to. You can generate this with: python3 gen_tone.py <path to MIDI file, should end with .mid> <path you want the tone outputted to, should end with .mp3>
4. Record yourself singing along to the "tone file" generated from the MIDI file. Store this in a file somewhere on your computer and take note of its location.
5. From the same directory as the cloned repo, run python3 gen_harmonies.py <path to MIDI file, should end with .mid> <path to you singing, should end with .mp3> <path you want the pitch-corrected version outputed to, should end with .mp3>

Samples:

• "The Longest Time" (Billy Joel)
• "Nobody Loves You Like Me" (Jonathan Coulton)
• "Auld Lang Syne" (traditional)
• "Doin' it Right" (Daft Punk)
• "Hide and Seek" (Imogen Heap)
• "Lift Every Voice And Sing" (James Weldon Johnson and J. Rosamond Johnson)
• "The Wellerman" (traditional, arr. The Longest Johns)

How Does This Work?

The algorithm used for pitch-shifting is called PSOLA, specifically Max Morrison's Python implementation of it. The program first reads a MIDI file and extracts the note times from it, then groups these into separate vocal tracks, and then pitch-corrects the MP3 file to each track using the PSOLA library. These harmonized tracks are then summed and the result is outputted to an MP3 file.