General

Interface preferences.

Theme Switch between light and dark interface. Saved instantly.

Connections

Service URLs for each backend. Change these first when setting up.

Local stack

TTS — Text to Speech Qwen3 engines: clone, design, custom, streaming · Kokoro FastAPI

Voice Clone / Base URL (WAV voices) Uploaded/cloned WAV voices. Expected: POST /v1/audio/speech.

Voice Design URL (instruction voices) Prompt-designed voices and vd_... virtual voices.

CustomVoice URL Style over configured speakers such as Ryan, Vivian, Serena.

Streaming URL Progressive low-latency WAV playback.

Kokoro FastAPI URL (82M · CPU-friendly) OpenAI-compatible TTS. Built-in voices: af_bella, bf_emma, am_adam… ~300 MB RAM.

VibeVoice URL (vibevoice-service) Simple text-in audio-out TTS. POST /tts with {"text":"..."}.

NVIDIA speech stack router, Magpie TTS, Parakeet ASR, clone NIM

NVIDIA router URL (TTS + STT) OpenAI-compatible base URL for the NVIDIA speech router.

NVIDIA Magpie TTS URL Direct Magpie endpoint. Fixed speakers, not WAV cloning.

NVIDIA Parakeet ASR URL Direct Parakeet endpoint for transcription.

NVIDIA Zeroshot NIM URL Magpie Zeroshot clone endpoint with audio_prompt.

NVIDIA Flow NIM URL Magpie Flow clone endpoint with audio_prompt and transcript.

STT — Speech to Text Whisper, faster-whisper, whisper.cpp, Groq, Parakeet, NVIDIA router

Active STT URL Default recognition endpoint. Expected: POST /v1/audio/transcriptions.

faster-whisper-server URL (CTranslate2 GPU) GPU-accelerated Whisper via CTranslate2. ~70× RT · 1.5 GB VRAM.

whisper.cpp URL (CPU/CUDA) Lightweight C++ Whisper server. ~8–15× RT CPU · ~1 GB RAM.

Groq API key (Groq Whisper & LLM)

Used for Groq Whisper STT and LLM. Free: 2 000 req/day.

Playback

Controls how previews play and how OpenAI-compatible requests are shaped.

TTS preview playback Buffered keeps Save WAV available. Streaming starts sooner.

OpenAI-compatible request style Controls payload shape for the normal TTS API URL.

Captures

Default behaviour for STT transcription, LLM text refinement, and playback.

Transcription Default language and preferred STT backend

Default language Language hint passed to the STT backend. Auto-detect works well in most cases.

Preferred STT backend Overrides the active STT URL for STT-TTS panel captures.

LLM server Base URL for persona rewrite, text refinement, and other local LLM features

LLM base URL OpenAI-compatible endpoint. Used for persona rewrites, voice design, and auto-refine.

LLM refinement defaults Automatic text cleanup after transcription

Auto-refine after transcription When on, the LLM refinement runs immediately after each capture.

Refinement model Model name sent to the LLM URL. Leave empty to use the app's current default.

Remove filler words Strips um, uh, like, you know, sort of.

Remove repetitions Collapses repeated words and phrases.

Fix grammar & spelling Light corrections without changing meaning.

Fix punctuation Adds missing commas, periods, and sentence casing.

Default playback voice Pre-selected voice in the STT-TTS panel

Default voice Pre-selects this voice in the STT-TTS panel on load.

Payloads

Extra JSON fields sent alongside each backend request. Usually only needed for advanced tuning.

Voice Clone / Base params Extra fields for the 8020 WAV voice clone/base model.

Streaming params Extra fields for the 8023 streaming model.

CustomVoice params Extra fields for the CustomVoice backend.

Voice Design params Extra fields for Voice Design and virtual vd_... voices.

NVIDIA Magpie params Usually empty. Magpie accepts fixed speakers: sofia, aria, jason, leo, john.

NVIDIA Zeroshot params Optional multipart fields. The app supplies text, language, audio_prompt.

NVIDIA Flow params Optional multipart fields. The app also sends the saved reference transcript.

Kokoro FastAPI params Extra JSON fields for Kokoro. Usually empty — voice is selected from the dropdown.

VibeVoice params Extra JSON fields for VibeVoice. Usually empty — service only requires text.

Storage

Container paths for voice files. Match these to your Portainer volume mounts.

Voice scan directory Contains active_voices, hidden_voices, sounds, and metadata.

Active voices directory New cloned/exported voices are saved here.

API Keys

Usually empty for local containers. Set a dummy value like sk-local if a backend requires an Authorization header.

TTS API key (optional)

Voice Design API key (optional)

Whisper API key (optional)

Backup & restore

Export all voices and settings as a ZIP for backup or migration. API keys are excluded from exports.

Export voices Import voices

Logs

Recent server activity. Useful for debugging backend connections and API errors.

Click Refresh to load logs

About

TTS Voice Creator

Clone, design, and deploy custom voices using local AI backends. Compatible with Qwen3-TTS, Kokoro FastAPI, NVIDIA Magpie, and any OpenAI-compatible TTS/STT endpoint.

FastAPI Vanilla JS WaveSurfer.js v7 MDI v7.4.47

MCP Server Expose voice tools to Claude Code, Cursor, and other MCP-aware agents

HTTP transport URL http://localhost:7890/mcp Streamable HTTP (JSON-RPC 2.0). Tools: speak, transcribe, list_captures, list_profiles.

Claude Code one-liner

claude mcp add voice-creator --transport http --url http://localhost:7890/mcp --header "X-Voice-Creator-Client-Id: claude-code"

Any HTTP MCP client (JSON)

{"mcpServers":{"voice-creator":{"url":"http://localhost:7890/mcp","headers":{"X-Voice-Creator-Client-Id":"my-agent"}}}}

/speak REST endpoint Generate audio from any app or script without routing rules

Generate speech (example)

curl -X POST http://localhost:7890/speak \
  -H "Content-Type: application/json" \
  -H "X-Voice-Creator-Client-Id: my-script" \
  -d '{"text":"Hello world","voice":"EN_F_Anna"}' \
  --output speech.wav

Per-client voice binding

curl -X PUT http://localhost:7890/speak/bindings/my-script \
  -H "Content-Type: application/json" \
  -d '{"voice":"EN_F_Anna"}'

Global hotkey daemon Push-to-talk transcription that types into any focused window

Install & run (host machine)

pip install pynput sounddevice soundfile pyperclip requests
python hotkey_daemon.py --server http://localhost:7890

Hold Ctrl+Shift+Space to record, release to transcribe and type. Linux: install xdotool for direct key injection.

Settings