TTS

The voice that reads tips on stream - queue, history, voice settings, and AI voices.

The TTS page runs everything about the voice on your stream: the queue of messages waiting to play, the history of what’s played, and the voice itself.

Queue and history

The Queue panel shows what’s waiting, in order. The History panel shows what’s already played - tips and alerts (subs, gift subs, resubs, cheers, raids, Kicks, channel-point alerts) in the order they happened - with a replay button on each, great for “wait, what did that say?” moments. The chips in the panel header narrow it to tips or alerts (every platform event, follows included).

An alert reaches the history once it has played on stream; one that was cut short from Controls says Skipped. An alert that is still waiting for an overlay, or that your tip filter blocked, is not listed.

Replayed tips and alerts show a gold “(Replay)” badge on the overlay so chat knows it’s a rerun. An alert replays through the alert box exactly like the first time, so the overlay you have open decides: if its alert box has that kind of alert switched off, or the cheer is under its bits minimum, the dashboard tells you instead of replaying into nothing. No overlay open? The replay waits, plays when one connects, and lands in the history then. A replay never counts twice - no sub-reward credit, no goal or label bump, no chat announcement, no webhook.

The voice

The voice lives on your alert box, per overlay. A tip’s audio is made at the moment the alert box plays it, in that box’s voice, so two overlays can sound different and nothing about the voice is set anywhere else. Open the Alert box widget in the overlay editor and find Voice:

• Voice - one of the standard voices (British, American, Australian, Indian, German, French, Spanish, Portuguese, Japanese, Korean - male and female where available), with a Preview button so you can hear it before a tip does.
• Speed and Pitch - how fast and how high it speaks.
• Read aloud on this overlay - the master switch for every spoken line here, tips and alerts.

What of a tip is shown and read - the donor’s message on the overlay, and whether the name, the amount and the message are read aloud, with how much of a long message is read - is on the Tip row under Alert events, the same way each event row has its own show and read toggles. The Tip row’s TTS switch is tips only. Volume sits with the other levels at the top of the widget.

Until you change any of these, each overlay keeps reading tips the way it does today.

Click a queued tip and press Preview TTS to make its audio right away, in your main alert box’s voice, so it is ready before the tip reaches the front of the queue. (Your main alert box is the one on your oldest overlay.)

Viewers get a Preview TTS button on your tip page too, so they can hear their message before paying - “Name tipped amount, message”, in your main alert box’s voice.

AI voices

AI voices are in beta: they need TipPage+. The AI voices tab on the TTS page says so if they aren’t on your account - check that your subscription is active.

An AI-voice tip is made once, at payment, in the donor’s chosen voices, and every overlay plays that one clip - so its opening line is set here, not per overlay. Spoken intro on the AI voices tab holds the wording (blank = the built-in “Name tipped amount”; {name} and {amount} are filled in) and whether the name, the amount and the message are read.

When it’s on, viewers can have their tip read out in an AI voice instead of the standard one - and they can write a script, where different lines of the message are spoken by different voices, played back-to-back as one clip.

On the tip page, viewers see a “Read it in AI voices” button under the message box. It opens a script editor: each line has a voice attached, tapping the voice name opens a picker with a playable sample for every voice, and pressing Enter starts the next line for a back-and-forth. The whole thing is voiced in one go - the model performs the exchange, including the timing between turns - with up to five different voices in a message.

Delivery tags

Under the script editor there’s an Advanced row: Delivery tags. These are short directions a viewer can drop into a line to tell a voice how to say the next bit - [excited], [whispering], [sarcastic] - or to make a noise instead of a word: [laughing], [sigh], [gasp], [pause]. Tapping a chip inserts the tag where the viewer is typing, and typing one by hand works too (case and spacing don’t matter).

A tag applies from where it sits until the next tag or the end of the line, so [whispering] don't tell anyone [shouting] I WON changes delivery mid-sentence. [emphasis] stresses the word right after it.

The tags on offer:

• Emotion - excited, happy, sad, angry, scared, surprised, disappointed, confused, proud, nervous, sarcastic, calm, confident, worried, frustrated, embarrassed, disgusted, grateful, curious, bored, hopeful, determined.
• Delivery - whispering, shouting, screaming, singing, speaking quickly, speaking slowly, soft tone, serious tone, low voice, emphasis, pause, short pause, long pause.
• Sounds - laughing, chuckling, giggle, sigh, gasp, inhale, exhale, crying, sobbing, groaning, panting, yawning, clear throat.

[singing] is the playful one: the model takes it as a natural-language cue rather than a built-in marker, so how well a voice sings varies by voice and by line. If it doesn’t take, the line is read normally.

Tags are direction, not speech. They aren’t read out, they never appear in the tip’s message on your overlay, in chat or in your history, and they don’t count towards the viewer’s character limit. There’s a cap of six per message, and the model does best with no more than three on one line.

The list of tags is fixed and we curate it - viewers can’t invent their own. Anything else typed in square brackets is treated as ordinary words and read aloud like the rest of the message, so it still goes through your word filter exactly like anything else a viewer writes.

Your controls in the panel:

• Offer AI voices to donors - the master switch.
• Mode - Donors can choose keeps the standard voice as the default, with AI voices as an option. All TTS uses AI voices removes the standard voice entirely: every tip message is AI-voiced.
• Minimum tip for AI voices - set it equal to your minimum tip for no extra requirement, or higher to make AI voices a premium that viewers unlock by tipping more.
• Voices - the voices viewers can choose from, with a play button to sample each one. Manage voices opens the two lists: This channel (voices you cloned, plus any you added from the library) and TipPage voices (ours, on by default - switch off the ones you don’t want, or all of them).

Manual tips in AI voices

Add manual tip on the Controls page has an AI voices switch next to the message field. Flip it and the message becomes the same script editor viewers get - speaker lines, a voice for each, delivery tags - so you can test a voice on stream or read something out in character. It works whether or not Offer AI voices to donors is on: that switch decides what your tip page shows viewers, not what you can put in your own queue. The voices on offer are the ones in your library, and a manual tip in AI voices is still a manual tip - no money, never counted in your totals.

Your own voices

Manage voices on the AI voices panel opens your voice library, where you can clone a voice from a recording:

1. Add a voice, give it a photo, a name and a short description (the name and description are what viewers see in the picker). The photo is square - pick any image and we crop and shrink it for you - and it’s how you tell your voices apart in the library.
2. Add up to 90 seconds of audio, in as many as ten clips. Clean, single-speaker speech works best: no music, no background chatter, 10 to 30 seconds at the energy you want. Shorter is also friendlier for back-and-forth messages - every voice in a message shares one budget, so a couple of 90-second voices can’t be cast together.
3. Optionally type what’s said in each clip. Leave it blank and we’ll transcribe it for you.
4. Tick the rights confirmation - it has to be your own voice or one you have permission to use - and hit Create voice.

Your uploads are deleted as soon as the voice is built - the voice itself is what we keep, and it lives on our own machines. Building takes a minute or two; the row says Building until it’s ready. Once it is, we record a short sample in the new voice (“Hi, this is a preview of my voice…”) and store it - that’s the clip you and your viewers hear when you tap play in the picker, so auditioning a voice is instant and never waits on the AI. Once a voice is built it’s finished - there’s nothing to tune or re-run. A failed build tells you why and offers Try again.

Voices start private to your channel. There are two ways to let other streamers use one:

• Share shows the voice’s code - a vc_… string minted with the voice that never changes. Give it to whoever you like (a friend, your Discord, a pinned message) and anyone in the beta can enter it on their own voices page to add the voice. The voice stays unlisted; only people with the code can find it.
• Publish lists the voice in the shared voice library so any streamer can find it. It doesn’t appear on their tip pages unless they choose to add it. Unpublish takes it off the list again.

Sharing and publishing are independent, and neither takes a voice away from anyone: a channel that added your voice - from the library or by code - keeps it whether you publish or unpublish afterwards, until they remove it or you delete the voice. Deleting is the only way to take a voice back, so share the code where you mean to.

Browse opens the library: everything other streamers have published, each with a sample to play. Who published a voice isn’t shown - only the name, photo and description its owner gave it. Add the ones you want and they join This channel; Remove takes one back out without affecting anyone else. Enter a code is the same thing for a voice someone shared with you directly. TipPage’s own voices aren’t in the library - they’re already on, in their own list, and you switch them off rather than adding them.

On your tip page, viewers see the same split: This channel for your voices and the ones you added, Other voices for TipPage’s.

Impersonation is the one hard line: we remove cloned voices that get reported as someone who didn’t agree to it.

A few things to know:

• Free messages can use AI voices too - but only while you haven’t set a higher minimum tip for them. A free message can’t clear a paid gate.
• Your word filter and AI tip filter still apply. If a filter changes the message (or a moderator edits it in review), the AI voices speak the filtered text, each line still in the voice the donor picked. Lines the filter rewrote lose their delivery directions (whispering, shouting and so on); lines it didn’t touch keep them.
• AI audio takes a little longer to generate than standard TTS, especially multi-voice scripts. The overlay holds the tip at the front of the queue until its audio is ready rather than playing it in the wrong voice.
• Deleting a voice that a waiting tip used means that tip gets read in your standard voice instead.
• Anything that stops the AI audio being made - a deleted voice, a busy GPU - falls back to your standard voice. The alert still plays.
• There’s no message preview for viewers, but every voice has a sample clip they can play before choosing.
• Automating it? The developer API can read the catalog and toggle the feature or individual voices.

Mod control from chat

Your mods don’t need dashboard access to run the queue - !tts skip, !tts pause, !tts resume, and !tts replay all work from chat once the bot is enabled.