Reading messages aloud
Text-to-speech: the three engines, what each costs, and how to pick a voice.
Any message can be played out loud. Tap the speaker button on it and the character's voice reads the reply; press and hold that same button for half a second and the voice picker opens instead.
The three engines
Which engine a voice belongs to decides what it costs and whether you need a plan for it.
| Engine | Voices | Cost | Subscription | Notes |
|---|---|---|---|---|
| Microsoft Edge | 60 voices across 22 locales | Free | No | The default, and free at any length. |
| MiniMax | 160 voices | 25 credits per character of text | Required | Richer expressiveness. The Premium badge in the picker. |
| Xiaomi MiMo | 9 voices | Free during a limited-time promotion (30 credits per character once it ends) | No | The Limited-time free badge. Also the only engine that can design a voice. |
If you never touch the setting, a character speaks with an Edge voice chosen from your language and the character's gender.
MiniMax is billed per character, not per message. At 25 credits per character, a 300-character reply costs 7,500 credits the first time it is played. The audio is then stored, so replaying it is free — for you and for anyone who plays it in a chat you shared. The first time you play a MiniMax voice, Reverie shows a Text-to-Speech Credits notice with the exact cost of that message; it appears once and then plays go straight through.
Without an active subscription, a MiniMax voice only plays audio that was already generated. Edge and MiMo have no plan requirement.
A shared chat plays the character in the voice its sharer chose, unless that voice is the sharer's own private clone; then it uses the character's own voice, or the default voice if the character has none.
Choosing a voice
A voice is stored per character, for you, and follows you across devices. Setting it does not change what anyone else hears.
Open the picker in either of two ways:
- Chat Settings → Voice & Input → Voice
- Press and hold the speaker button on any message
The picker has two tabs — Preset voices and My voices (your own clones and designs from the Voice Studio). Filter by language and gender, or search by name or style. Voices are grouped as Character voice, Recently used, For this character and All voices, and each carries a price badge: Free, Limited-time free or Premium.
Preview plays a sample line in your language before you commit. A custom voice previews from the clip Reverie generated when the voice was created.
The voice a creator chose
A character's creator can set a default voice in the character editor, and you hear that voice unless you pick another. If the default is a cloned voice, you can preview and use it on that character but you cannot add it to your own library.
Playing replies automatically
Auto-play Voice plays each new reply as it arrives. The toggle is in Chat Settings → Voice & Input, and it only appears while Instant Chat Mode is on — it is the setting for that mode, not a global one. Its scope is This device.
Voice calls ignore it entirely: a call always speaks.
What actually gets spoken
By default the speaker button reads only the dialogue: the words inside quotation marks, or, with the Speech with (actions) format, the text left once the parenthesised actions are dropped. Narration, scene descriptions and actions are skipped, and a reply with no dialogue in it can't be played.
Read Whole Message reads the whole reply instead, narration and actions included; code blocks and tables, such as status panels, are still skipped. The toggle is in Chat Settings → Voice & Input and its scope is This device. It is hidden while Instant Chat Mode is on, because replies in that mode are written to be spoken and are always read in full. A whole reply is longer than its dialogue, so a paid engine charges more for it.
Before anything is synthesised, Reverie strips sticker and emoji-pack images, markdown formatting and emoji out of the text, so the voice reads the line rather than the punctuation around it.
Each synthesis request is capped at 500 characters after that cleaning. Longer messages are split into chunks of up to 480 characters and played back to back, so a long reply still plays in full. The cost is unchanged — a paid engine bills per character either way.
Voice notes a character sends
A character can send a real voice note on its own instead of a text message, up to 200 characters, in one of seven emotions. Replaying a voice note you were sent is free — Reverie verifies the message and skips the charge. See Proactive messages.
Related
- Voice calls — a live spoken conversation instead of playback
- Voice Studio — make a voice that isn't in the preset list
- Credits — where the per-character charge lands
- Subscriptions — what unlocks the MiniMax engine