It’s time for a throwback – this video was produced in 2014. But don’t miss out on a serious vocoder love fest full of history and celebrity interviews. And just wait, because the vocoder may be about to make an AI-fueled comeback.

The New Yorker talks to the likes of Laurie Anderson, Cozmo D, Dave Tompkins, and Frank Gentge about what made this instrument special, and traces its weird, twisted history through military applications to Kraftwerk parties.

The topic of vocal encoding and the vocoder has become freshly relevant with the rise of machine learning for vocal synthesis. The “AI” trend is driven in no small part by a resurgent vocoder – only this time, powered by neural networks (themselves a revived technology, after a long “winter”). These vocoders use neural networks to “learn” and process, enabled by massively parallel computation on GPUs and specialized “AI” chips:

https://gfx.cs.princeton.edu/pubs/Jin_2018_FAR/

https://github.com/candlewill/RawNet

… just to give two examples. The expected application of most of this is text-to-speech (TTS). Think talking translators and speaking apps and so on, just more futuristic (or nightmarish, depending on your feelings about that futurism).

But as always with the vocoder, musical applications do have a way of holding their own. Right now, it’s computationally expensive to train vocoders. But it is possible to cheat, by doing most of the pre-training work and then letting a user do some light training at the end. That is, you probably sound similar to other people speaking your native language, so it’s possible to train a machine on those details rather than make it start from scratch.

What I’m getting at: it’s almost inevitable that we’ll soon see musical vocoders and pitch correction that is trained on your voice. And in turn, that could create presets that abuse different vocal characters for various creative impacts. (Maybe there’s something like this now, I just haven’t seen it in the market.)

Whether or not that proves useful, understanding the history of the vocoder means getting a deeper grasp of how communications technology has evolved generally – and how people can push its envelope to make something expressive. Whatever useful applications folks like military leaders may imagine, us humans do love to be human and push the emotional boundaries of the tech we touch.

(Just of course “no one needs a vocoder” – Robert Henke. Okay, he was joking.)