Releases
Changelog
What's new in each Cantai installer release.
Cantai now runs natively on Intel Macs. Download the dedicated macOS (Intel) installer from the downloads page — the Apple Silicon installer won't install on an Intel Mac, and vice versa. Rendering on Intel runs on the CPU, so it is slower than on Apple Silicon; we recommend at least 4 CPU cores and 8 GB of RAM.
- Cantai now supports Intel Macs: a dedicated installer for Intel-based Macs (macOS 13.4 Ventura or later) renders locally, just like on Apple Silicon.
- Cantai now tells you when an update is available: a banner in the plugin links straight to the new installer, and an outdated engine shows a clear update prompt instead of “Not Connected”.
- More pronunciation fixes: “mazed”, “wildered”, “beheld”, “sapling”, “cobble” and the elision “’neath” now sing their correct vowels and syllables.
- Fixed a rare crash that could happen during playback while a render was running.
- On Windows, zooming with Ctrl+scroll no longer breaks the plugin window.
We recommend updating Sibelius to version 2026.8 — together with this release, it makes editing a score with Cantai active noticeably smoother.
- Renders are noticeably faster: the heaviest parts of the voice engine now run on your GPU when one is available, on both macOS and Windows, falling back to the CPU automatically.
- Editing a few bars no longer re-renders the whole score — unchanged sections are reused from a cache, and a small edit now re-renders only the moments around it instead of everything that follows.
- Sibelius stays responsive while you edit: quick successive edits, batch operations like Explode, and mute or solo changes no longer set off waves of unnecessary re-renders in the background.
- Clearer diction on unstressed syllables: words like “infinity” now sing their real vowels instead of an “uh” sound — and dozens more words were tuned individually, including “tomorrow”, “forget”, “begin”, “either”, “pharaoh”, and words ending in “-le” like “little” and “double”.
- Crescendos and diminuendos now build gradually across their whole length, the way a singer would phrase them, instead of only stepping at the arrival — and loudness no longer jumps between rendered sections.
- The singers' own phrasing now carries through to the final mix: each phrase keeps its natural rise and fall, and the overall level is set once for the whole part instead of being flattened section by section.
- Voices no longer sing through short written rests, even at slow tempos.
- “Clear caches…” now lets you clear just one part, the current score, or everything — and cleared audio stops playing immediately.
- Dorico no longer freezes or crashes when closing a project.
- A redesigned controls panel and a more polished installer on both platforms.
For now, choose your singers in the Cantai panel rather than in the Sibelius Mixer. Open the Cantai plugin window for a stave, pick the voice there, and it stays on that stave and is saved with the score. The Mixer lists the general voice types — Soprano, Alto, Tenor, Baritone, Bass — and the choir sounds. Choir and ensemble sounds assigned in your score are unaffected, and existing scores keep playing: a stave that named a specific singer falls back to its voice type, and you can set the singer again in the panel. This is a provisional solution — we are working on the definitive one, and we will ship further improvements as they are ready. If something still looks off, tell us through the chat on cantai.app; the more you can describe about the score, the faster we can reproduce it.
- Written dynamics now shape the singing itself — a crescendo or a forte adds real vocal effort and brighter tone, not just more volume, and hairpins ramp toward the next marking. Scores with no written dynamics sound exactly as before.
- Steadier pitch — noticeably less drifting out of tune on long held notes and long vowels, and far fewer of the sudden dives into a low growl at the start or end of a note. This is an area we are still actively improving.
- Cleaner sound — removed the clicks at the seams between rendered sections, and fixed a timing wobble that made small ensembles beat against each other.
- Clearer diction — corrected six English words (amen, forever, retain, present, Israel, Hallelujah), smoothed Spanish diphthongs into a single vocal gesture, and fixed phrase-opening consonants arriving too early.
- Sibelius: staves no longer end up without a voice or swap voices with each other after a Mixer change, “No parts found” now recovers on its own, and a render that never starts no longer leaves the plugin stuck on “Rendering…”.
- Sibelius: the “Bass” sound now loads the Ottetto bass voice (Patrick) instead of a baritone, so a Bass stave sings in the right range.
- Sibelius: for now, singers are chosen from the Voice selector in the Cantai panel, one per stave — the pick is saved with the score and stays on its stave, and the Mixer lists the general voice types and the choir sounds. This is a provisional solution while we work on the definitive one.
- Dorico: the “Auto & Cantai” playback template now ships with the installer, so Cantai is assigned to every singer automatically — and the plugin window no longer shrinks into the top-left corner on high-resolution displays.
- Dorico: fixed renders that could lose their dynamics, or fail outright, when Dorico cleaned up a file Cantai was still reading.
- The score language is now saved with the score, so different scores can use different languages and changing one no longer changes the others.
- Closing one notation app no longer stops rendering in another one you still have open.
- One installer for everyone — Dorico and Sibelius now ship in a single installer.
- Added Spanish lyrics support, with a language selector right on the main screen.
- Added the Ottetto ensembles — ready-made choral groups: Sopranos, Altos, Tenors, Basses, Women's Voices, Men's Voices, and Full Ensemble.
- Added six new solo voices: Eric, Sergey, Aggie, Rachel, Leah, and Jesse.
- Refreshed the plugin interface to match the new Cantai look, with Reverb and Voices controls on the main screen.
- Reverb settings are now remembered per stave in Sibelius.
- Fixed smaller issues: voice changes arriving too early, a newly added track not rendering, and the Settings button occasionally not responding.
- Corrected pronunciation of 'into', 'this', 'Mar', and 'the' in sung lyrics.
- Sibelius: voices now load reliably even when a voice is selected before the list finishes loading.
- Sibelius: fixed a crash that could occur during automatic rendering.
- Sibelius: an outdated version now shows a dismissible warning instead of blocking rendering.
- Solo voices now use a subtler, warmer hall reverb.
Dorico and Sibelius now share a single installer and release together.
The releases below are from before the merge, when each app shipped separately.
Earlier releases
- Added Reverb and voice controls directly on the main render screen
- Improved render quality — clearer sound on loud passages and fewer detuning artefacts on long notes
- Tuned pitch expressiveness for a more stable pitch
- Fixed wrong pitch or speed when the project's sample rate changed
- Fixed the plugin freezing when the rendering engine was unreachable
- Added the new Patrick voice, now in the Bass/Baritone section
- Added three new voices: James, Kim, and Maxwell
- Voices are now organized into Soloists and Ensembles banks in the voice selector
- Added an Export Logs button in Settings for easier troubleshooting
- Fixed notes without lyrics rendering as silence — they now sing a placeholder vowel
- Fixed notes being sung longer than they should
- Fixed a crash that could occur while a render was starting
- Fixed activation and deactivation screen flickering on Windows
- Fixed words with curly/smart apostrophes (e.g. "don't", "it's") being skipped or misaligned to notes
- Fixed humming passages ("mmm"-style notes) not rendering correctly
- Fixed certain common English words being mispronounced when sung
- Fixed the first note being cut off when it starts at the very beginning of the timeline
- Fixed the plugin window appearing blurry on high-DPI Windows displays
- Fixed a macOS "open application" dialog that could pop up inside the plugin UI
- Improved server status detection so the plugin no longer reports the server as down while a render is in progress
- Fixed an issue where rendering could get stuck and never complete
- Fixed lyrics occasionally being assigned to the wrong notes
- Fixed words ending in -ing being sung as -ung
- Fixed words with apostrophes (e.g. "don't", "it's") being skipped or mispronounced
- Fixed rendering freezing on Windows in certain situations
- Moved the Jon voice to the Tenor group
- Fixed missing singer photos on Windows
- Fixed a rare issue where two simultaneous renders could corrupt each other's audio
- Added free trial — new users can try CantAI before purchasing
- Fixed the plugin occasionally showing the wrong subscription status
- Fixed a crash when lyrics contained certain punctuation characters
- Renamed the Aji voice (previously listed as "Aji JP") and corrected her voice type to Alto
- Rendering now starts playing back sooner — audio comes in progressively rather than all at once
- Singer photos now appear in the voice selector
- The CantAI engine now launches automatically when you open your DAW on macOS
- The plugin and the CantAI engine are now bundled in a single installer
- Fixed incorrect timing on rests and pauses
- Fixed a crash that occurred after rendering completed on Windows
- Added new voices
- First Windows release
- Choral rendering now works fully offline
- Initial release of the CantAI plugin for Dorico