Voice Over

What this page is

Voice → Voice Over reads donations, chat and bot replies out loud. The line under the title says where the sound comes from: "Synthesis runs on this machine — no internet, no limits".

How to set it up

The page has four tabs — Voice, What to speak, Rules, Sound in OBS — and one Save button at the top right for all of them. Above the tabs runs a live strip: a dot with Silent or Speaking now, In queue: {count} underneath while lines are waiting, and two buttons — Skip applies to the line being read, Stop and clear to everything at once.

Voice

1. The block at the top of the tab is the voice pack. On a fresh install it reads Neural voices are not installed — "One download of {size}. Until then the system voice of Windows is used — it works everywhere but sounds noticeably worse." Press Download voices. That opens the terms first — which model the voices come from, under which licence and what it forbids. Accept and download starts the download, Read the licence opens the full text in a browser, Cancel closes the screen.
2. While it runs the title becomes Downloading… {percent}% with a progress bar and the file being fetched; Cancel stops it. When it finishes the block reads Neural voices are ready — "{size} on disk. Ten voices, 31 languages, everything stays on this machine" — and the button becomes Remove, which asks first: "Remove the voice pack? Voice over will fall back to the system voice, and the cached audio will be cleared. Settings are kept." If a download was interrupted the title reads The voice pack is incomplete; downloading again replaces the incomplete files.
3. Voices — Neural voices or System voices of Windows.
4. Language — All languages plus every language the pack covers. The hint explains the point of the list: "Pick a voice for every language your viewers write in. The language of the text decides which voice speaks — the system voice needs no download but sounds robotic." Pick the language first, then the voice for it, and repeat for the next language.
5. Under the list are the voice cards, named Male 1, Female 1 and so on. Click a card to choose that voice for the language currently selected; the play button on the card is Listen. Pressing it before the pack is downloaded reports Download the voice pack first.
6. With System voices of Windows the language list and the cards disappear and a note takes their place: Windows picks the voice itself — "It reads with the first voice installed for the language of the text — there is nothing to choose here. Download the voice pack to pick a voice and get a far better one."
7. In the Sound section: Speech rate — a slider from 0.70 to 1.50, 1.05 by default — and Volume, 80% by default. The hint below: "Volume is multiplied by the «Voice (TTS)» slider in «Settings → Sound». Beyond this rate range the model starts swallowing syllables."
8. Test phrase — a field ("Type anything and press Speak") and the Speak button; Enter in the field does the same. "Both the listen buttons and this one read exactly this phrase. Write it in the language you want to hear." The phrase is kept by Save.

What to speak

Four blocks, each with a switch. Under the name of a block it says Reads out loud or Off, and a block that is off hides its own settings.

1. Donations — on out of the box. Minimum amount: "Compared in your base currency, so a donation in another currency is judged correctly. Zero reads every donation." Template starts as "{user} donated {amount} {currency} and says: {message}", and the variables it accepts are listed right under the field: {user}, {amount}, {currency}, {message}. Which sources count is a checkbox per installed donation connector; these checkboxes take effect on their own, without Save.
2. Chat — off out of the box. Read is either On command only or Every message ("Every message turns into noise on a busy chat — that mode is for small streams"), and Command holds the command itself, !tts by default. Pause between lines is in seconds, 3 by default: "Seconds, counted per viewer. Donations and subscriptions are never held back." Template starts as "{user} says: {message}" and takes {user} and {message}.
3. Bot replies — on out of the box: "Whatever the AI agent answers by voice. Turning the voice on for the agent itself stays in its own settings."
4. Subscriptions — off out of the box. Subscription template starts as "{user} subscribed for {months} months" and takes {user}, {months}, {tier}; Gifted subs template starts as "{user} gifted {count} subscriptions" and takes {user}, {count}, {tier}.

Rules

1. Length — Length limit in characters, 300 by default ("Characters. Zero means no limit"), and When it is too long: Trim it or Skip the whole message.
2. Filters — Cut links out, on by default: "Otherwise a viewer pays to have the bot read a website address into your stream." Blocked words, one per line: "A message containing any of these is not spoken at all. Whole words, not parts of them." Blocked viewers, one nickname per line.
3. Queue — Queue limit, 10 by default: "When the queue is full, chat is dropped first — a donation never pushes out another donation. Anything dropped is written to the log."

Sound in OBS

The tab is the instruction itself, under How to get the voice on stream: "In OBS: Sources → + → Application Audio Capture, and pick strAImer in the window list", then "Set the volume in the app: Settings → Sound, slider Voice (TTS)", then "Press Test phrase on the «Voice» tab and watch the source meter in the OBS mixer". One source covers everything the app plays — voice, alerts and scenario sounds go through it.

The second section, The voice overlay is no longer needed, asks you to delete the browser source of the voice overlay: it no longer plays anything, because "OBS puts a browser source to sleep in silence and loses the beginning of the audio, which is why playback moved into the app". The last section, If it stays silent, lists the three things to look at: the capture source in the OBS mixer, the volume in Settings → Sound, and the state strip at the top of this page.

What it cannot do

• The Listen buttons and Speak read only what stands in Test phrase — nothing else can be previewed here.
• System voices of Windows gives no choice of voice: neither the language list nor the voice cards are shown in that mode. The cards belong to the downloaded pack.
• The pack arrives as a single download. Individual voices or languages cannot be installed separately, and Remove takes the whole thing away.
• Speech rate cannot go outside the slider's range, and the volume here is not the final volume: it is multiplied by Voice (TTS) in Settings → Sound, which is a different page.
• The queue is not shown line by line and cannot be reordered — the strip gives the count, and the only two actions are Skip and Stop and clear.
• Every message reads what viewers say, not what they type as a command. A chat line starting with an exclamation mark is sorted as a command the moment it arrives from Twitch, YouTube or Kick, and no command reaches this mode — not even !tts itself. Commands are spoken only under On command only, and only the one written in Command. Nothing here makes Every message cover them as well.
• Which sources count exists only under Donations. Chat, bot replies and subscriptions have no source filter of their own.
• Whether the AI agent answers by voice at all is not decided here — that stays in the agent's own settings.
• Voice no longer plays through a browser source, so there is no overlay link on this page.