audio8bit 0.0.0

Turn a song into an 8-bit / chiptune arrangement of its melody (vocal or instrumental)

audio8bit

Turn any song into 8-bit, video-game-style music - right from your terminal. audio8bit finds the song's melody (and its chords) and replays them with retro "chiptune" sounds, like an old game console.

English | Español | Português | Français | Deutsch | Italiano | Русский | 中文 | 日本語 | हिन्दी | 한국어

What it does

• Give it a song, get back a chiptune version of it.
• Works whether the song has singing or is instrumental - it picks the tune automatically.
• Everything runs on your own computer; nothing is uploaded.

Before you start

You need two things:

• Python 3.9 or newer
• ffmpeg - a free tool for reading and writing audio. Install it with sudo apt install ffmpeg (Linux) or brew install ffmpeg (macOS).

Install

pip install audio8bit

First run is slow: it downloads a small AI model (about 80 MB) and can take a few minutes. That's normal - later runs are faster.

Use it

audio8bit -i song.mp3

This creates output.mp3 in the current folder. That's it. Each run also prints a short quality report so you can see the result came out clean.

Want something different? Here are the most common tweaks:

audio8bit -i song.mp3 -V lead          # just the main melody, no chords
audio8bit -i song.mp3 -s vocals        # follow the singing
audio8bit -i song.mp3 -s instrumental  # follow the instruments
audio8bit -i song.mp3 --transpose 5    # play it 5 semitones higher
audio8bit -i song.mp3 -f ogg           # save as .ogg instead of .mp3

All options

Option	Default	What it does
-i, --input	required	The song to convert (mp3, wav, flac, ...)
-o, --output	output.<type>	Where to save the result
-f, --format	same as input	Save as a different type, e.g. ogg, wav
-s, --source	auto	Where to take the tune: vocals, instrumental, or auto
-m, --method	transcribe	How notes are found: transcribe (best) or pitch (faster, lighter)
-V, --voices	chords	chords (with harmony) or lead (one melody line)
--transpose	0	Shift the key, in semitones (e.g. 5 up, -5 down)
--bits	8	Sound resolution, 1-8 (lower = crunchier)
--rate	22050	Sample rate in Hz (lower = more retro)
--duty	0.25	Tone colour of the pulse wave, 0-1

If something goes wrong

• "ffmpeg not found" - install ffmpeg (see Before you start).
• The first run seems stuck - it's downloading the AI model; give it a few minutes. It only happens once.
• It doesn't sound like the song - try -s vocals or -s instrumental to pick the right part, or -V lead for just the melody.

How it works (optional reading)

1. Splits the song into parts (vocals, drums, bass, and the rest).
2. Detects the actual notes being played in the part you chose.
3. Replays those notes with simple 8-bit "chip" sounds and saves the file.

License

Noncommercial