TTS Connectors

TTS Connectors

Text-to-Speech Setup Options.

New
ddistro pockettts
PocketTTS
http://127.0.0.1:8086
0 assignments
New TTS Connector (Copy) (Copy)
PocketTTS
http://127.0.0.1:8086
0 assignments
New TTS Connector (Copy) (Copy) (Copy) (Copy)
PocketTTS
http://127.0.0.1:8086
0 assignments
New TTS Connector (Copy) (Copy) (Copy) (Copy) (Copy) (Copy)
PocketTTS
http://127.0.0.1:8086
0 assignments
New TTS Connector (Copy) (Copy) (Copy) (Copy) (Copy) (Copy) (Copy)
PocketTTS
http://127.0.0.1:8086
0 assignments
OmniVoice Default
OmniVoice
http://127.0.0.1:8021
0 assignments
Russian Piper Irina (local)
Piper TTS
http://127.0.0.1:18766
1 assignment
Clone Delete
Please save any changes before testing to ensure the latest settings are used.
This label appears in profile and player connector pickers.
The provider driver this connector loads at runtime.
Used for providers that expose a local or remote HTTP endpoint.

NPC Fallbacks

NPC male fallback VoiceID if the NPC voice is blank or the provider rejects it.
NPC female fallback VoiceID if the NPC voice is blank or the provider rejects it.

OmniVoice Settings

No OmniVoice language profiles were found in /home/dwemer/omnivoice-tts/languages.

Higgs TTS 3 Settings

Model ID exposed by the Higgs service. Default: higgs-v3.

PocketTTS Settings

audio.cpp model id. Default: pocket-tts.

Chatterbox Settings

Enable paralinguistic tags like [laugh], [sigh] for expressive TTS output.
Prompt snippet instructing the LLM to use paralinguistic tags. Added to system prompt when enabled.
Comma-separated list of supported paralinguistic tags (e.g., [laugh],[sigh],[gasp]). Tags are case-insensitive.

CHIM XTTS Settings

Enable paralinguistic tags like [laugh], [sigh] for expressive TTS output.
Prompt snippet instructing the LLM to use paralinguistic tags. Added to system prompt when enabled.
Comma-separated list of supported paralinguistic tags (e.g., [laugh],[sigh],[gasp]). Tags are case-insensitive.

Inworld TTS Settings

Inworld workspace ID (required for voice cloning). Format: workspaces/{workspace} or just the workspace ID.
Language to use for TTS generation. Uses BCP-47 regional language codes.
Inworld model to use. inworld-tts-2 prioritizes quality; inworld-tts-2-flash prioritizes speed.
Sampling temperature (0-2). Higher values make output more random. Default: 1.0
Speaking rate/speed (0.5-1.5). Default: 1.0
Let the LLM write delivery and sound tags for Inworld, e.g. [smug] Nice swing. [laugh]. With inworld-tts-2 the LLM may also write its own short directions such as [mad as hell]. Tags are removed from subtitles and history. Edit tags

Cartesia TTS Settings

Language to use for TTS generation. Available languages depend on the selected model.
Choose Sonic 3.5 or 3.6, or a dated snapshot to pin a release. Existing connectors keep their selected model.
Optional, Sonic 3.6 only. For multilingual voices, enter an accent ID supported by the selected voice, such as general-american or standard-hindi. Leave blank for the voice default.
Speaking speed for the voice

Piper-TTS Settings

speaking time scale. Use a value over 1.0 to play slower, a value under 1.0 is faster.
speaking variability. Leave 0 to use voice model internal value. Experiment with values around 0.667
phoneme width variability. Leave 0 to use voice model internal value. Experiment with values around 0.8
Name of speaker for multi-speaker voices (if you have an .onnx voice file with multiple voices).
Speaker id for multi-speaker voices, overrides speaker name if used. 0 is default (first) speaker.

xVASynth Settings

Base language
Model Type
xVASynth version (e.g. 3.0 is default. Older models are 1.0 or 2.0)
xVASynth gameID (e.g. skyrim)
Pace
Wave Glow Path (relative)
Vocoder
Leave as default!

Azure Text-To-Speech Settings

Force mood (voice style)
Region location of your API key
Volume
Talk speed
Voice contour
Allowed voice styles

ElevenLabs Text-To-Speech Settings

Reduces response delay at some cost to quality and text handling. Use 0 for default behavior; higher values favor speed more aggressively.
ElevenLabs model to use for this connector. Use eleven_v3 if you want V3 enhancers or audio tags.
Controls how consistent each generation sounds. Lower values are more expressive, while higher values are steadier and less varied.
Controls how closely the output sticks to the selected voice. Higher values usually sound more like the source voice, but can reduce flexibility.
Adds extra stylization and exaggeration to the delivery. Higher values can sound more dramatic, but may increase latency.
Adjusts the speaking rate of the generated audio. 1.0 is normal speed.
Boosts resemblance to the original voice at a small latency cost. Ignored by eleven_v3.
Controls whether numbers, dates, abbreviations, and similar text are rewritten before speech. Auto lets ElevenLabs decide, on always normalizes, and off keeps the raw text.
Enables extra language-specific cleanup before synthesis. This is mainly useful for languages that need special text handling, such as Japanese.
Optional Eleven v3 prompt tags added before the text, such as [whispers] or [curious]. Only used when model_id is eleven_v3.

OpenAI TTS Settings

Model
Control the voice of your generated audio with additional instructions. Does not work with tts-1 or tts-1-hd.

KOKORO 82M Settings

Speed

KoboldCPP TTS Settings

This TTS provider does not have any connector-level settings to configure here.

Zonos TTS Settings

Language
Default profile only. Model to use.
Default profile only. Enhance emotional quality by requesting values from the LLM. If false, emotions will be determined by the LLM-selected mood.
Pitch standard deviation [0-300]
Speaking rate. Higher is faster. [5-30]
CFG scale. Controls how closely the audio matches the sample voice. Higher numbers will be a closer match. [1.1 - 5]
Path to the sample audio stored in zonos. Leave as default; it is set automatically.

Deepgram TTS Settings

Model