You just talk
Speech recognition runs continuously while the call is live — no record button, no walkie-talkie pauses.
A real two-way voice conversation with any character — they listen while they speak, you can cut them off mid-sentence, and the next line comes about a second after you stop talking.
Mordred Pendragon
Why voice
Text is a great place to write a scene. Voice is where a character starts to feel like company.
— Reverie team
One turn of conversation
No push-to-talk, no waiting for a recording to upload. The call keeps its own rhythm.
Speech recognition runs continuously while the call is live — no record button, no walkie-talkie pauses.
When you stop talking, the character takes the turn — typically about a second later, like a natural pause in conversation.
The reply streams as audio while it's being generated, in the character's own voice, so you're never staring at a spinner.
Start speaking while the character is mid-sentence and they stop to listen — exactly like cutting in on a real call.
The voice studio
Pick from 60+ voices across English, Chinese, Japanese, Korean and more — then tune until it sounds like them.
Nudge the pitch, slow the delivery from 0.5× to 2×, and save the result to that character. Your setting sticks for future calls.
Call a whole scene. Each character in a group chat speaks with their own voice, and the active speaker lights up.
What you get
The details that make voice feel like presence instead of a feature demo.
Listening and speaking run at the same time. The character hears you while they talk — which is what makes interrupting possible at all.
Silence detection hands the turn over almost immediately. No dead air, no tapping a button to signal you're done.
English, Chinese, Japanese, Korean and more, drawn from two TTS engines — enough range to cast any character believably.
Voice, pitch, and speed are saved per character, so the detective never accidentally sounds like the princess.
Multi-character scenes work on voice too. Everyone keeps their own voice, and the speaker is always highlighted.
The credit rate is shown on the call screen before and during the call. No hidden voice surcharge — you see exactly what a minute costs.
A sample of the library — every one of them can pick up.
Common questions
Plenty of apps bolt a play button onto a chatbot and call it voice. A Reverie voice call is a live, interruptible, two-way conversation with the character — closer to a phone call than to an audiobook.
Your speech is transcribed as you talk, using browser speech recognition or Whisper-class transcription — no push-to-talk, no manual send step.
Responses are synthesized and streamed in chunks, so the character starts speaking while the rest of the reply is still being written.
Speak over the character and playback stops immediately while the system listens to you instead. Conversations can overlap, the way real ones do.
Roughly a second of silence hands the floor to the character. There's no awkward gap where you wonder whether the app heard you.
A voice call is only as good as the voice on the other end — and only as trustworthy as its bill.
Two voice engines provide a deep library across English, Mandarin, Japanese, Korean and more, so characters can speak the language of your roleplay.
Each character keeps their own voice, pitch, and speed settings. Set it once and every future call sounds like them.
Multi-character conversations carry over to voice: each cast member speaks with a distinct voice and the active speaker is highlighted on screen.
Voice calls are metered per minute in credits, with the rate displayed before you connect and pinned to the call screen while you talk.
When you're ready
Pick a character, press call, and talk. Per-minute pricing stays on the screen the whole time.