Voices

What this page is

Voice → Voices is where you pick who speaks with which voice. Everything in the app that reads text out loud asks this page for a voice: donations, chat, subscriptions, channel point rewards, the bot's own lines in chat, the assistant's answers and scenarios.

One voice for everything used to be the only option, and it lived inside the Voice Over settings. It moved here for two reasons. Different things deserve different voices — a donation and your assistant answering a question are not the same event. And the assistant should be able to talk even if you never installed the Voice Over set: the machinery that turns text into sound has always been part of the app itself, not of that set.

The voice pack

The card at the top is the neural voice pack: ten voices that speak 31 languages, synthesised on your own computer. It is one download of about 380 MB and it stays on your machine — no text ever leaves it.

Until you download it, the app reads with the Windows system voice. That works everywhere and sounds noticeably worse. Press Download voices, read the short licence summary and accept it. The download takes minutes; you can leave the page while it runs.

Remove deletes the pack and the cached audio. Your settings are kept, and the app falls back to the system voice.

Roles

Below the pack is one row per task that asks for a voice. The list is built from what you have installed, so it is short if you installed little:

• Donations, Chat, Subscriptions, Channel point rewards and Bot in chat appear when the Voice Over set is on.
• Bot replies appears when the AI agent is set up.
• Scenarios appears when the Scenarios set is on.

Two of those rows are about the bot, and they are different things. Bot in chat is everything the app writes to chat under the bot's name — a command reply, a timer, an announcement. Bot replies is what the assistant answers *you* out loud. Give them the same voice if you think of the bot as one character, or different voices if you want to hear which of the two is talking.

Pick a voice from the dropdown next to a role and press Listen — the app reads a test phrase the same way it will read it on stream: same voice, same speed, same volume, through the same output. If you hear nothing, the problem is in the sound path, not in this page; see the guide "Sound".

Default voice means "decide for me". It is a valid choice, not an empty one — the app picks a voice and reads.

A voice for one language only

If you set up voices per language in the old Voice Over screen, those settings came across and show as extra rows underneath the role. The × next to such a row drops it, and that language goes back to the role's main voice.

You will not see these rows otherwise, and you do not need them: every voice in the pack speaks all 31 languages. "Male 3" is not a Russian voice — it is a person who also speaks Russian.

How it speaks

The bottom card holds the settings that belong to the machine rather than to any one task, because you have one set of speakers:

• Voice — neural, system, or off. Off means the app says nothing at all, whatever the roles say.
• Speed — 0.7 to 1.5. Past that range the model starts swallowing syllables.
• Volume — multiplied by the Voice (TTS) slider in Settings → Sound.
• Queue limit — how many lines wait their turn on a wave of events. When the queue is full, the cheapest line is dropped first: chat before subscriptions, subscriptions before donations, and a donation never pushes out another donation. Zero means no limit.

What is not here

Whether something gets read at all is not decided on this page. That stays where the rules live:

• donation threshold, phrase templates, blocked words and viewers, chat cooldown — Voice → Voice Over;
• whether the assistant answers out loud — Chat Bot → AI Agent, the *Read answers out loud* switch on the Actions tab.

This page answers "with which voice", not "whether".