VST FINDS ( )

Everybody Wants a Robot Choir

CRUMPETS and klattsch turned formants into the week's most shareable nerd sport on X. What each one actually does, in the words of the people who built them.

September 19, 2026

Some links on this page are affiliate links: if you buy through them, VST FINDS earns a commission. It never changes your price.

Every few months a plugin announcement escapes the gear-forum aquarium and swims out into the general feed. Usually it happens because somebody made a knob do something a knob has no business doing. This week it happened twice, and both times the trick was the same unglamorous organ of the human voice: the formant.

The breakfast pastry that splits your voice in two

CRUMPETS comes from CRQL. Ewan Bristow's September 12 post announcing it ("Our Cepstral formant shifter plugin. Now available") had pulled more than 46,000 views and 434 likes by the morning of the 19th. The name is a joke with a lecture folded inside it: CRUMPETS is an anagram of cepstrum, which CRQL's own page calls "the spectrum of a spectrum."

That phrase is doing a lot of work. A spectrum tells you which frequencies are in a sound. Take the spectrum of that, and the two things that make a voice a voice stop overlapping. One is the resonant shape of the throat and mouth, the formants. The other is the evenly spaced teeth of the harmonics riding on the pitch. By CRQL's account, CRUMPETS pulls those apart into two separate signals and shifts the formants. Then it stitches everything back together and returns it to the time domain. "How fun!" the product page adds, which may be the most honest sentence ever written about DSP.

The controls read like a lab bench with a sense of humor. Res sets the cepstral cutoff between formant and harmonic information. Mid and Side shift the formants separately, and Lock shifts them together. There is a Gate, a Mix, and an Auto mode that tracks the cutoff on monophonic material. There is also MIDI control of that cutoff, so you can play the seam between voice and pitch like an instrument. The display paints formants orange and harmonics blue, for anyone who wants to watch the surgery.

It ships as VST3, CLAP and AU for Windows, Mac and Linux, for £13 (about $17). The demos on its page run from vocals to neuro bass.

A 1980 research lineage that learned to sing

If CRUMPETS is the scalpel, klattsch is the whole robot. It started, in its maker's words, as a joke that "unexpectedly took off, with millions of views." Its developer, Tony Gies (on X as @me_irl), turned the afternoon project into a full app. The press kit is precise about the family tree: the synthesis "follows the Klatt formant-synthesis research lineage associated with classic computer voices." That line of work runs back to Dennis Klatt's 1980 paper on a cascade/parallel formant synthesizer. The kit is just as precise about what klattsch is not: it "does not contain DECtalk code and does not try to sound convincingly human. The artificial character is the point."

The feature list is a small love letter to the uncanny:

The page's own description of the result beats anything we could add: "strange, buzzy, and stiff, but eerily alive."

The full app is $9.99 on itch.io for Windows, macOS, Linux and Android. It is DRM-free and offline, with no ads and no accounts, and the browser version stays free. On September 19, Gies posted on X that people have been asking for a plugin, and that he would rather build the best formant speech-synthesis plugin than a quick and dirty one.

Why formants, why now

Put the two side by side and a small pattern shows up. Voice models can clone almost anyone now, yet the vocal tools that caught fire on X this week treat the voice as a mechanism to take apart. In klattsch's case, it proudly refuses to pass for human at all. That's not a trend report. It's two data points and an opinion. But it's a decent opinion, and both of these are worth a look.