Voice calls
Talk to characters out loud. Continuous flow, emotional voices, no push-to-talk.
Voice calls turn a Reverie character into something closer to a phone call than a chatbot. There's no push-to-talk; you just speak and stop, and the character speaks back. Their voice carries emotion — warmth, hesitation, glee — synthesised in real time.
Why we rebuilt voice in 2025: the old system felt transactional — press, wait, press again. The new one feels like a call. Read the rebuild story →
Starting a call
- Open any chat.
- Tap the 📞 call icon in the top right.
- Grant microphone permission (one-time, per browser).
- Start talking. The character speaks first if the conversation is new; otherwise they respond after you finish.
To end the call, tap the red hang-up button. Your call transcript appears in the chat as normal messages, so the conversation continues seamlessly when you switch back to typing.
How the conversation flows
- No push-to-talk. Start speaking whenever; the character listens, replies when you pause.
- Interrupt at any time. Start speaking over them and they'll stop and listen.
- Silence is okay. Long pauses don't end the call — the character waits.
Voices
Every character has a default voice picked by the creator. You can override it per character under Voice settings, reachable two ways:
- The gear icon during a call, or
- Long-press the speaker / play (🔊) button on any message in the chat.
You can also set the default voice and Auto-play Voice from the side panel's Advanced Settings → Voice.
| Engine | Languages | Notes |
|---|---|---|
| Microsoft Edge TTS | 17+ | Free. Clear, steady, no emotion. Excellent for everyday calls. |
| MiniMax Speech-2.8 | English, Chinese, Japanese, more | Premium. Emotion-aware: warmth, sadness, anger, excitement. Best for emotional scenes. |
Knobs you can tune per character: engine, voice ID, speed (0.85×–1.2×), pitch, and — on MiniMax — emotion preset.
Picking a voice that fits the character →
Pricing
| Service | Status |
|---|---|
| TTS (synthesising the character's speech) | Spends credits per character read aloud — visible in the credit ledger per call. |
| ASR (transcribing your speech) | Currently free during the promo period. |
Rates are tunable — we adjust them as voice models become cheaper or more expensive. The credit ledger always shows the actual debit per call.
Languages
Voice calls work in every language Reverie supports — characters reply in your interface language by default and pick up your language from your speech automatically. Japanese, Korean, Mandarin, and Cantonese all have native-quality voices on the MiniMax engine.
Privacy
Audio is processed only for the duration of the call and isn't stored. Transcripts persist in your chat history just like typed messages.
Tips
- Pick MiniMax for romance, drama, and storytelling. The emotion synthesis is the whole point.
- Pick Edge TTS for tutors and assistants. Clear, no-frills reading.
- Test the voice on the character profile before calling. Every profile has a "play sample" button on the voice picker.
- Use a real microphone. Laptop mics work fine, but a headset is a huge upgrade for long calls.