OpenWhisperer 2.0.0 — it speaks Dutch (and German, Polish, Russian, Ukrainian)
Open Whisperer can now speak languages Kokoro has no voice for. A second TTS engine, Supertonic-3, runs alongside Kokoro and takes the languages it can't.
Your Dutch reply is now actually Dutch
Before this, a Dutch reply was read aloud by an English voice. That's worse than it sounds — it isn't "Dutch with an accent", it's unintelligible. Feeding the synthesized audio to a speech recognizer and asking it to write down what it heard:
| what the recognizer heard | error rate | |
|---|---|---|
| Before (English voice) | "D test Slagin New Alemolol." |
~100% |
| After (Dutch voice) | the sentence, one word off | ~4% |
The original was "Ik heb de wijzigingen doorgevoerd in het configuratiebestand. De tests slagen nu allemaal…".
What's new
- Five new languages in Settings → Voice: Dutch, German, Polish, Russian, Ukrainian — two voices each. Pick one and replies are spoken in that language.
- Replies are written in your voice's language. Picking a Dutch voice isn't enough on its own — a Dutch voice reading English text helps nobody. Your agent is now told to write the spoken summary in that language. Your on-screen reply is unaffected.
- It's fast. The new engine runs on the Neural Engine at ~19× realtime — roughly 10× quicker than the existing English voices, so spoken replies start sooner.
- Nothing changes if you're on English. The multilingual model (~264 MB) downloads only when you actually select one of these voices. Existing voice settings keep working untouched.
Also in this release
- Settings window is slightly slimmer.
- Removed the redundant underline beneath the logo tab.
Notes
- Requires Apple Silicon, macOS 14+.
- Mandarin stays on the existing Kokoro voices — the new engine wasn't trained on it.
- Gatekeeper: as with previous releases, this build isn't notarized. On first launch, right-click the app → Open, then confirm.