TTS
Text-to-speech with local Piper voices, StreamElements, and ElevenLabs.
TTS
The TTS plugin speaks text through local Piper voices, StreamElements, or ElevenLabs.
Setup
Open Plugins → TTS → Configure and set up one or more providers:
| Provider | Requirements |
|---|---|
| Local TTS | Download Piper voices in the app |
| StreamElements | StreamElements overlay token |
| ElevenLabs | ElevenLabs API key |
Each provider is a panel you can open and close. While a panel is closed, a badge in its header shows the status, such as Connected, Not configured or Runtime required. All panels start closed.
ElevenLabs account widget
Add the ElevenLabs account widget to the dashboard to see:
- The linked account and its subscription tier
- Credits used, credits left and the credit limit, with a usage bar
- When your credits reset
- Used voice slots
Use Update API key in the widget to replace the key. Stream Kit checks the new key with ElevenLabs before saving it, and the settings page shows the new key right away. Restricted API keys need the User: Read permission for the widget.
ElevenLabs voices
Open TTS → ElevenLabs voices in the sidebar to manage your voices. The list shows 50 voices per page; search and the All voices / My voices filter cover every voice on the account.
- Preview: press play next to a voice to hear it. Voices without a hosted sample play a short test sentence instead.
- Tune: give a voice its own stability, similarity, style, speed and speaker boost. Use Test voice in the dialog to hear the result before saving, and Reset to default to go back to the global tuning. Tuned voices get a Tuned badge.
- Model: in the tuning dialog, pick the model this voice uses. The model chosen in an action wins; otherwise the voice's model is used, and then the default model from the TTS settings.
- Volume: turn on Custom volume in the tuning dialog to give a voice its own volume. Otherwise it uses the ElevenLabs volume from the TTS settings.
All TTS volume sliders (Local, StreamElements, ElevenLabs and per voice) go up to 200%. Above 100% the audio is amplified, which can clip on loud recordings.
- Audio tags: in the tuning dialog, add tags such as
[heavy german accent]or[whispers]. They are placed before every message spoken with that voice, but only when the model in use supports audio tags (Eleven v3 and newer). Pick such a model for the voice to make sure the tags are used. Older models skip them, so the tags are never read out loud. - Add voice: clone a voice from one or more audio samples (one to two minutes in total works best). Files over 10 MB are converted to mono WAV, trimmed to 5 minutes and split into parts automatically. ElevenLabs accepts up to 25 samples per voice. You must confirm you have the rights and consent to clone the voice. Optionally remove background noise.
- Delete: cloned, generated and professional voices can be removed from your ElevenLabs account. Deleting a voice in Stream Kit also deletes its local backup.
- Copy voice ID for use in actions. In TTS → ElevenLabs → Speak Text, choose Variable for voice and use
{voiceId}from context.
Choose a default model on the configure page; override per handler or use Use default model.
Per-voice tuning is stored in Stream Kit and synced with your Stream Kit account, not at ElevenLabs.
Voice backups
Cloned voices live in your ElevenLabs account, so they disappear from Stream Kit when you switch to an API key of another account. Backups keep them on this device:
- Automatic backup: every voice you clone in Stream Kit is saved on this device (the uploaded samples, name, description and noise-removal setting) under the app data folder. Turn off Keep a backup on this device in the clone dialog if you don't want that for a voice.
- Back up: cloned voices that were made elsewhere (for example on the ElevenLabs website) show a Back up button that downloads their samples from ElevenLabs. Backed-up voices get a Backed up badge. Only instant clones with samples can be backed up; generated, professional and library voices cannot.
- Automatic restore: when you save a different ElevenLabs API key, every backed-up voice that is missing from that account is cloned again from its samples. The voice gets a new voice id at ElevenLabs; its per-voice tuning, the default voice in the TTS settings and every Speak Text action that used the old id are moved to the new one, so nothing has to be set up again. Old ids that still arrive through variables keep working too.
- Switching back: when you return to an account that still has the voice, the backup is linked to that voice again by name, and tuning and actions follow.
- Restore by hand: backups that are not on the connected account (for example because the account has no free voice slots) are listed at the top of the voices page with Restore and Restore all buttons. Delete backup removes just the local copy; the voice at ElevenLabs stays.
Backups are stored per device and are not synced with your Stream Kit account. Restoring a voice uses a voice slot on the new account like any other clone.
ElevenLabs voice settings
Tune the default voice under ElevenLabs → Voice tuning on the configure page and use Test voice to hear the result before saving.
| Setting | Range | Default | Effect |
|---|---|---|---|
| Stability | 0–1 | 0.5 | Lower is more expressive, higher is more consistent |
| Similarity | 0–1 | 0.75 | How closely the output follows the original voice |
| Style exaggeration | 0–1 | 0 | Amplifies the speaker's style; higher values add latency |
| Speed | 0.7–1.2 | 1 | Speaking rate |
| Speaker boost | on/off | on | Boosts similarity to the original speaker |
These are the defaults for every voice. Tuning a voice on the ElevenLabs voices page overrides them for that voice. Settings are applied in this order: global defaults, then per-voice tuning, then per-action overrides.
TTS → ElevenLabs → Speak Text can override stability, similarity, style, and speed per action. Leave a field empty to use the default; variables from the trigger work too. Values outside the range are clamped, and anything that is not a number falls back to the default.
ElevenLabs privacy
Don't store messages and audio at ElevenLabs is on by default:
- Enterprise accounts use zero-retention mode (
enable_logging=false). ElevenLabs does not store the text or the generated audio. - Other accounts don't support zero retention. Stream Kit detects this once, then removes each message (text and audio) from your ElevenLabs history right after it is generated. For a short moment the item exists in your history. If removal fails you get one warning per session.
Turn the switch off if you want to keep TTS messages in your ElevenLabs history.
Handlers
| Handler | Description |
|---|---|
| TTS → Skip | Stop the currently playing TTS message (and discard in-flight synthesis); the next queued clip still plays |
| TTS → Local → Speak Text | Offline Piper voice |
| TTS → StreamElements → Speak Text | Cloud via StreamElements |
| TTS → ElevenLabs → Speak Text | Cloud via ElevenLabs |
Each handler supports a fixed voice or a Variable voice ID.
Speak handlers finish after their audio has finished playing, or after a playback timeout (about two minutes) if the audio device hangs. A stalled ElevenLabs request is aborted after 30 seconds. Failed ElevenLabs or StreamElements requests and missing voices show an error toast. In all these cases the handler fails softly and the next clip or action can run — the TTS queue will not stay blocked. Use an action queue with blocking enabled when later actions must wait for that speech to end.
Reference: TTS handlers.