Pick the key and scale, and every note is drawn to the nearest one in it. Drag the puck left for a hard, robotic snap and right for a slow, natural glide; up humanizes, leaving notes that are already close to their note alone, down corrects every one. The note follows the pitch as it is sung, so a wide vibrato swings to the next note and back: the classic hard-tuned sound. It works in real time: the host compensates the 0.25 ms of latency it reports, and while a voice is being tuned the sound runs further behind, 3.8 ms on average. It is not a harmoniser and not a vocoder: it corrects one voice at a time.
Read the manual for every control and how to use it.
What it's for
Features
- 13 scales in every key: Chromatic, Major, Natural, Harmonic and Melodic Minor, Dorian, Phrygian, Lydian, Mixolydian, Locrian, Major and Minor Pentatonic and Blues.
- One puck for the feel: SPEED from 0 to 200 ms along the pad, HUMANIZE from 0 to 100 % (up to 50 cents) up it. 0 ms snaps at once.
- The note follows the pitch: a wide vibrato swings to its neighbour and back, for the hard-tuned sound.
- 0.25 ms of latency in every voice range, reported to the host so the track stays in time. While a voice is being tuned the sound runs a little further behind, 3.8 ms on average.
- The voice range beside the scale: Low, Normal or High, each with its range in Hz.
- A display that shows what it does: the last four seconds, the voice as sung and as corrected, with STRENGTH, DETUNE and A4 set right in it.
- A jellyfish that listens: its bell pulses with your level and its tentacles stir the further you are off the note, calm within 5 cents.
- Light: 0.13 % of a CPU core at the defaults.
- Zoom from 100 to 200 %.
- No network access and no telemetry.
Install
Download the latest version from the Releases page, or with these commands:
curl -LO https://github.com/jelly-comet/JellyPitchBtw-release/releases/download/v1.3.0/JellyPitchBtw-1.3.0-linux-x86_64.tar.gz
curl -LO https://github.com/jelly-comet/JellyPitchBtw-release/releases/download/v1.3.0/JellyPitchBtw-1.3.0-linux-x86_64.tar.gz.sha256
sha256sum -c JellyPitchBtw-1.3.0-linux-x86_64.tar.gz.sha256 && tar xzf JellyPitchBtw-1.3.0-linux-x86_64.tar.gz
cd JellyPitchBtw-1.3.0-linux-x86_64
Install for your user:
mkdir -p ~/.clap
rm -f ~/.clap/JellyPitchBtw.clap && cp JellyPitchBtw.clap ~/.clap/
Or for all users:
sudo mkdir -p /usr/lib/clap
sudo rm -f /usr/lib/clap/JellyPitchBtw.clap && sudo cp JellyPitchBtw.clap /usr/lib/clap/
Removing the old copy first avoids overwriting a file a running host has loaded. Then rescan plugins in your DAW.
To uninstall:
rm -f ~/.clap/JellyPitchBtw.clap
FAQ
Is it free?
Yes. The download is free and MIT-licensed.
Is there a VST3 or LV2 version?
No. Releases contain the CLAP plugin only.
Does it run on ARM, like a Raspberry Pi or Apple Silicon?
No. It needs Linux on x86-64, on a CPU with AVX2.
Which hosts does it work in?
Any CLAP host. It was tried in Bitwig Studio and REAPER.
Does it work on Wayland?
Through XWayland. The editor also needs a Vulkan driver.
Does it work on chords or several voices at once?
No. It corrects one voice at a time: a lead vocal, a backing vocal, or an instrument that plays one note at a time. It is not a harmoniser and not a vocoder.
What's new in v1.3.0
14 Nov 2026 · Release on GitHub
JellyPitchBtw 1.3.0: a quarter of the latency.
- 0.25 ms of latency in every voice range, reported to the host, where 1.2 had 1 ms.
- The sound a little nearer: while a voice is being tuned it runs 3.8 ms behind on average, where it ran 4.5 ms. What is left is the voice's own cycle, which the plugin repeats or drops now and then to move a note: about 2 to 3 ms for a high voice and 5 to 6 ms for a low one.
- The same correction and the same CPU as 1.2: 0.13 % of a core at the defaults. The output is not the same sample for sample, and a corrected note lands a touch later against the sound: at SPEED 0 ms a voice sits a median 6 cents from its note, where 1.2 left 5.
A project saved with 1.2 opens with every setting.
Linux x86-64, CLAP. Install: see the README.
Earlier versions
v1.2.0 14 Nov 2026
JellyPitchBtw 1.2.0: less latency, lighter, and corrections that land sooner.
- 1 ms of latency in every voice range, reported to the host. It was 3.75 to 10.75 ms, 6.81 ms at Normal. While a voice is being tuned the sound runs a little further behind, 4.5 ms on average where it ran 6.7 ms.
- Corrections land sooner: the pitch detector reads the newest part of what it has heard, 4.5 to 9.5 ms sooner at Normal. At SPEED 0 ms a corrected voice sits a median 5 cents from its note, where 1.1 left 9.
- Lighter: 0.13 % of a CPU core at the defaults where 1.1 took 0.26 %, and about a fifth of the memory per instance.
- It needs a CPU with AVX2 now (Intel Core from 2013, AMD from 2015). On an older CPU, stay with 1.1.0.
- The sound is 1.1's hard-tuned correction. To sit that near the input the voice is joined to itself in a new way, so the output is no longer the same sample for sample.
A project saved with 1.1 opens with every setting. VOICE no longer changes the latency.
Linux x86-64, CLAP. Install: see the README.
v1.1.0 14 Nov 2026
JellyPitchBtw 1.1.0: the classic hard-tuned sound, simpler.
- The note follows the pitch: a wide vibrato swings to the next note and back, the sound of a hard-tuned vocal. The correction is the one VocalTuneBtw had.
- SPEED from 0 ms: all the way left snaps to the note at once.
- HUMANIZE replaces VIBRATO up the pad: 0 to 100 % of 50 cents, how far off its note a voice may be and be left alone. It starts at 0 %.
- One latency: 3.75 to 10.75 ms by voice range, 6.81 ms at Normal (48 kHz), reported to the host. REALTIME and the Quality look-ahead are gone.
- VOICE beside the key and scale, with a wave icon and the range in Hz for Low, Normal and High.
- FORMANT and HOLD are gone; the display's top row is STRENGTH, DETUNE and A4.
- 0.26 % of a CPU core at the defaults.
A project saved with 1.0 opens with its key, scale, SPEED, STRENGTH, DETUNE, A4 and VOICE; REALTIME, VIBRATO, FORMANT and HOLD are dropped, and HUMANIZE starts at 0 %.
Linux x86-64, CLAP. Install: see the README.
v1.0.0 13 Nov 2026
JellyPitchBtw 1.0: pitch correction for one voice, steered by a jellyfish.
- One puck for the feel: SPEED, 1 to 200 ms, along the pad and VIBRATO, 0 to 100 %, up it. Bottom left is the hard, robotic snap; top right a correction you barely hear.
- The note through a vibrato: the note is chosen from the centre of what is sung, so a wide vibrato stays on its note.
- 13 scales in every key, Chromatic to Blues.
- REALTIME, 0.25 ms, for singing through it, or Quality, a look-ahead of 16 to 43.5 ms that puts each correction on the start of its note, for mixing and bouncing. Both are reported to the host.
- A display of the last four seconds, the voice as sung and as corrected, with STRENGTH, DETUNE, FORMANT, HOLD, A4 and VOICE set right in it.
- A jellyfish that listens: its bell pulses with the level, its tentacles stir the further the voice is off its note.
- 0.13 % of a CPU core at the defaults.
License
MIT, see LICENSE. The licenses of the third-party code in the binaries are in
THIRD-PARTY-LICENSES.txt in each release archive.
VST is a registered trademark of Steinberg Media Technologies GmbH. CLAP is an open standard by the free-audio project. JellyPitchBtw is an independent project, not affiliated with or endorsed by the makers of any other pitch correction.






