TTS providers
Every text-to-speech provider, what it needs, and which controls it supports
Vfinity can speak with six providers: one free, five that bill your own account. Choose one under Provider in SettingsAudioVoice.
Providers
| Provider | Needs | Models | Speed | Pitch |
|---|---|---|---|---|
| Microsoft Edge (free) | Nothing | — | Yes | Yes, in Hz |
| ElevenLabs | API key, Voice | Multilingual v2 (default), Turbo v2.5, Flash v2.5, Eleven v4, Eleven v4 Turbo, Eleven v3 | Yes, except Eleven v3 and v4 | No |
| Speechify | API key, Voice | Simba 3.0 (default), Simba 3.2 | No | No |
| Murf AI | API key, Voice | No model picker | Yes | Yes, on Murf's own scale |
| Cartesia | API key, Voice | Sonic 3 (default), Sonic 2 | Yes | No |
| Fish Audio | API key, Voice | S2.1 Pro (default), S2 Pro, S1 | Yes | No |
A new setup starts on Microsoft Edge (free) with the voice
en-US-AriaNeural. Every other provider speaks nothing until it has an API key
and a voice; the settings say so, and the voice list loads once the key is in.
ElevenLabs' Eleven v3 and v4 models have no speed setting, so Vfinity sends them none and Speed leaves them as they are.
Which voices laugh
With Read chat laughs as a laugh on, a laugh typed in chat is sent to the voice as a laugh. Whether it sounds like a real laugh depends on the provider and model.
| Provider | How it laughs |
|---|---|
| Fish Audio | A real laugh sound, with every model |
| Cartesia | A real laugh sound with Sonic 3; Sonic 2 says the laugh |
| ElevenLabs | A real laugh sound with Eleven v3, Eleven v4 and Eleven v4 Turbo; the older models say the laugh |
| Microsoft Edge (free), Speechify, Murf AI | Says the laugh, such as haha or jaja, in the voice's language where it is known |
The Laughs section in Chat TTS names how your voice laughs, and points you to Eleven v3, v4 or Sonic 3 when the model you have only says it. Text to speech covers the laugh styles.
Fish Audio's voices
Fish Audio lists your own cloned and private voices first, under Your voices, and searches its whole catalogue by name. Custom voice ID takes a voice model ID that the list does not show.
Provider settings
| Setting | Range | Notes |
|---|---|---|
| Provider | The six above | Each provider keeps its own key, voice and model, so switching loses nothing |
| API key | Your key from the provider's site | Never included in a backup |
| Voice | From the provider's list | Refresh the voice list reads it again |
| Model ID | A suggestion, or any model ID up to 200 characters | Empty uses the provider's default |
| Speed | −50% to +100% | Set for chat in Chat TTS and for Events lines in the Voice card. Speechify has no speed control, and the settings say so |
| Pitch | −50 to 50 | Microsoft Edge (free) and Murf AI only; others say This provider has no pitch control. |
Voices given to single viewers, and the random pool new viewers draw from, are kept per provider too, since each provider has its own voice IDs. A voice ID can be pasted straight into the pool.
What a paid provider bills
Paid providers bill per character. These keep the count down:
- Alerts read aloud count, as well as chat. Read alerts aloud is off by default.
- A Speak a custom line step stops once its event has spent its Characters per minute, 600 by default.
- With Play TTS in set to the OBS audio source and no such source open in OBS, chat and alerts are not read at all.
- Max characters cuts a long message before it is sent, 200 by default.
When a provider fails
| Message | Means |
|---|---|
| TTS is not set up | The provider still needs its API key or a voice |
| TTS provider rejected the request | Check the API key, voice and model |
| TTS provider quota reached | The provider reported a usage or rate limit; speech stays silent until it clears |
| Text-to-speech is unreachable | No connection to the provider; TTS resumes on its own |