Text-to-Speech (TTS) lets streamion.bot read alert text aloud through your audio output. It pairs with the alert overlays — follow, sub, cheer, raid, and other event toasts can be both displayed visually in OBS and narrated by a synthetic voice.

The feature is available to every workspace whose operator has registered at least one Piper voice model on the platform (Piper is the open-source speech engine streamion.bot uses to generate the audio). There is no Pro entitlement required.

Page surface

There are two TTS surfaces in the dashboard:

• Per-alert Voice card on /alerts/configure/<event-type> — choose whether each event speaks, override the voice, and decide which parts of the alert are read aloud. See the Voice section on the Alerts page.
• Workspace defaults on /settings/tts — pick the workspace-wide default voice that every alert without an explicit override uses, and set the maximum message length that gets sent to synthesis.

Prerequisites

• The overlays.manage permission. By default this is granted to the Owner, every Admin, and every Moderator.
• At least one Piper voice model registered by your operator on /admin/piper-models. If no models have been registered, both the TTS settings page and the Voice card on the alert configuration page show an empty-state message that points to the admin path; the controls activate as soon as a voice is available.

Default voice

Every alert that does not explicitly pick its own voice falls back to the default voice configured here. The dropdown lists every Piper voice currently registered on the platform; the labels include the language and speaker name supplied by the model.

Max message length

The slider limits how many characters of a viewer-supplied message (Resubscription text, Cheer message) are sent to the speech engine. Messages longer than the limit are cut off before they are read out. This setting controls how long a single TTS alert can speak and acts as a guardrail against spam from very long chat messages.

Voice catalogue

Voices are platform-provided rather than per-workspace. New voices appear in the dropdown the moment the operator registers them; voices that are removed from the platform fall back to the default for any alert that previously selected them.