There’s an old adage in software. If something you wrote isn’t selling, open source it.
Facebook/Meta Labs open sourced, AudioCraft, its family of AI audio models. The family consists of AudioGen, MusicGen, and EnCodec. AudioCraft is used for generating sound effects, audio tracks, and encoding audio for playback.
You can think of AudioCraft as a musical GPT. You prompt AudioCraft with a phrase and AudioCraft generates sound effects or music based upon your prompt.
Pop dance track with catchy melodies, tropical percussions, and upbeat rhythms, perfect for the beach.
MusicGen Prompt Example
Generative Audio Fails to Catch On
Unlike ChatGPT and Bard, generative audio AI hasn’t caught on. Lots of projects were started yet failed to gain traction. OpenAI, of ChatGPT fame, launched Jukebox to lukewarm usage. Google announced MusicLM last January, and the same folks behind Stable Diffusion launched Riffusion.
Music generation projects just haven’t attracted the talent and attention image and text generation projects have. Maybe, just maybe, there’s a difference in the sensory perception needed to make good music.
Songs and ideas that are musical can be mathematically produced. Most humans can intuit when music is offbeat, off pitch, or is weirdly composed. Most humans also know ‘good music’ when they hear it.
Good music stirs something different within us. Music’s vibrations can stir our soul as well as our ears. Music by itself can make us emotional. We can feel sadness while hearing melancholy chords, and the right song can get us moving in the morning for a hard day at work.
AI Can’t Makeup Something New
Generative AI is derivative. Whenever I think of generative music AI, I’m reminded of the scene in Back to the Future. The scene when Chuck Berry’s cousin calls him and tells Chuck about the new sound he’s looking for.
Music is beyond mathematics. It’s more.
Other than playing copycat using rappers’ voices and tracks to remix something, music may be a medium the nerds don’t fully understand how to synthesize. And that’s not to say there aren’t music playing nerds. I’m one of them.
Maybe, just maybe, feeling music is what makes us human and not machines.