Voice over: donations and chat read out loud

Everything it will read, and this is the whole list

Nothing is read aloud until you switch it on. Every source has its own switch and its own phrase template.

• Donations from an amount you set — the threshold is counted in your base currency, so a donation in another currency is judged correctly.
• Chat: on the “!tts” command only, or every message — the second one is for small streams.
• Subscriptions and gifted subs, each with its own phrase.
• Channel point rewards — the message a viewer typed together with the reward. Either every reward that carries a message, or only the ones you tick.
• What the bot writes in chat — AI agent answers, replies to commands, timers, scenario messages, stream announcements. Five separate switches, so replies can stay text while the assistant speaks, or the other way round.
• Chat arrives from Twitch, YouTube and Kick, donations from whichever services you connected — and you pick which of them count, because Twitch bits rarely need reading.
• The phrase is built from a template, so “Sasha sent 500 and says…” reads the way you want it to.

What it will not say

The rules are the whole point. Reading everything is unbearable within twenty minutes.

• A length limit that either trims the message or skips it whole.
• Links cut out — otherwise a viewer pays to have a website address read into your stream.
• Blocked words: a message with one of them in it is not spoken at all. Blocked viewers, by nickname.
• A pause between chat lines, counted per viewer — donations and subscriptions are never held back by it.
• A queue keeps a burst of donations from talking over itself, and when it fills up chat is dropped first: a donation never pushes out another donation.

Not a cloud service with a meter running

Cloud voice over is billed per character and stops working when the service is down or the plan runs out. This one is a file on your disk.

• The voice pack is 384 MB, downloaded once, and after that nothing goes over the network.
• No per-character cost, so a long donation message is not a decision you have to make.
• Ten voices and 31 languages in one pack: one voice speaks all of them, and every language can still be given its own.
• Until the pack is downloaded it speaks with the Windows system voice — noticeably worse, but working from the first minute.
• The download opens with the licence of the model and what it forbids: pretending to be a real person, passing synthesized speech off as a real recording, and harassment, deception or disinformation.

How the parts split up

The synthesis itself lives in the core; this module only decides what and when. The split is deliberate — several modules need a voice, but the rules for when to stay quiet are different for every streamer.

• The engine decides what to read and queues it; the app plays the audio itself.
• No browser source to add: in OBS one Application Audio Capture source pointed at strAImer carries the voice, and the alerts and scenario sounds with it.
• The module has its own volume slider, and it is multiplied by the “Voice (TTS)” slider in Settings → Sound.

Stopping it mid-word

Sooner or later a viewer sends something you would rather the stream did not hear. That is what the panel tile is for.

• The tile shows what is being read now and the queue behind it — each entry named by its kind and the viewer it came from, not by the text itself.
• “Skip” drops the current message, “Stop” silences the whole queue.
• Both are one press away while you are playing.
• The same two controls sit at the top of the settings page, where “Stop” is spelled out as “Stop and clear”.

What it is made of

Each part is a separate module: you install only the ones you need, and any of them can be removed without breaking the rest.

Getting started

1. Open Voice → Voices and download the voice pack.
2. Switch on what should be read and set the amount donations start from.
3. In OBS add one Application Audio Capture source and pick strAImer in it — this is how the voice reaches your viewers.
4. Put the Voice over: Queue tile on the control panel, within reach.

Voice over never calls the AI itself: the phrase comes from your own template and the voice is made on this machine — the agent only adds one more line to read.

Works together with