Features
Voice
Two separate things live under "voice": dictating your own messages instead of typing them, and having a companion's replies read aloud. Dictation needs no key and no setup. Reading aloud needs a provider key and is off until you turn it on.
Dictating a message
Tap the microphone in the message box and start speaking. Your words are transcribed into the input field — tap again to stop. The transcription lands in the box like typed text, so you can edit it before sending, or add to it.
Nothing needs configuring. There is no key, no provider to choose, and no setting to enable.
Dictation by platform
Uses the browser's own speech recognition. Works in Chrome, Edge and Safari; Firefox doesn't implement it, so the microphone won't work there — use another browser or the mobile app.
Uses your device's own speech recognition — the same engine your keyboard uses. Always available, and nothing leaves the device for the transcription itself.
Where the browser has no speech engine, Ensoul has a fallback: it sends the recording to a hosted transcription service instead. That path needs an OpenAI key, and the app doesn't accept one yet — so where the browser can't transcribe, dictation currently can't work there either. The desktop app is the same client in a native shell, so it depends on an engine being available inside it. In practice, dictation is most dependable in a Chromium browser or on Android.
Compare with Platforms for how this stacks up against the rest of the app's platform differences.
Reading replies aloud
Off by default. Turn it on in Settings → Voice (on the web, that's the Voice section inside Models, keys & APIs) and choose a provider, and each reply can be spoken rather than read — either automatically, or on demand with the speak control on a message.
ElevenLabs is the provider that works today. Paste a key, and you can load the voices on your account and pick one. The OpenAI voice options are present in the interface but can't be brought into a working state yet, because the app doesn't accept an OpenAI key — and even once it does, OpenAI's voice endpoints can't be called from a browser tab, so that path would be desktop-only.
Each companion can be given their own voice in their own settings, so a group conversation doesn't sound like one person arguing with themselves.
If it isn't working
- The microphone does nothing: the browser or app needs microphone permission — check the site or app permissions, and your operating system's privacy settings.
- No microphone button in Firefox: expected. Firefox doesn't implement the speech recognition Ensoul uses. Try Chrome, Edge or Safari.
- The transcription is inaccurate: it comes from your browser or device's own speech engine, not from Ensoul, so quality tracks that engine. Background noise and distance from the microphone matter more than anything else.
- Replies aren't spoken: reading aloud is off by default, and needs an ElevenLabs key. Check Settings → Voice (inside Models, keys & APIs on the web) — and note that a companion only reads aloud if you've turned it on for them too.