TTS Chat (LM Studio + Kokoro)

This is a modular Flask application that provides a modern chat interface powered by an LM Studio backend. It seamlessly integrates with Kokoro to synthesize high-quality text-to-speech for AI replies. The web/ frontend communicates with a flexible backend API that provides text generation, model discovery, and speech synthesis.

Requirements

• Python 3.10+ and below 3.14 with pip
• LM Studio running locally with a chat model loaded and the API server enabled (defaults to http://127.0.0.1:1234/v1 and key lm-studio)

Quickstart

• Installation & Running:

  python -m venv .venv
  source .venv/bin/activate  # On Windows, use `.venv\Scripts\activate`
  pip install --upgrade pip
  pip install -r requirements.txt
  
  # Start LM Studio separately, then run the app:
  python run.py

  Then open http://localhost:5000 in your browser to use the chat UI.

Config (Arguments & Environment Variables)

All settings can be passed as command-line arguments or environment variables. Arguments take precedence.

Argument	Env Variable	Description	Default
--base-url	LMSTUDIO_BASE_URL	Base URL for the LM Studio API	http://127.0.0.1:1234/v1
--api-key	LMSTUDIO_API_KEY	API key for LM Studio	lm-studio
--model-name	LMSTUDIO_MODEL_NAME	Force a specific model ID	(auto-detect)
--repo-id	KOKORO_REPO_ID	HF repo for Kokoro weights	hexgrad/Kokoro-82M
--lang	KOKORO_LANG	Default language code	a
--voice	KOKORO_VOICE	Default voice name	af_nicole
--sample-rate	KOKORO_SAMPLE_RATE	Output sample rate	24000
--port	PORT	Server port	5000

Example using arguments:

python run.py --port 5001 --voice af_bella

Voice catalog

Available voices are parsed directly from app/services/VOICES.md. Each entry exposes a name and lang_code to the frontend dropdown. You can update that file to add or remove voices without touching any system code.

Utility Scripts

I included a standalone script in scripts/terminal_chat_stream.py which allows you to chat interactively with your local model using your terminal. This script automatically streams spoken audio line-by-line using Kokoro as the LLM generates responses.

How it works

• /api/models proxies LM Studio to list available model IDs.
• /api/chat sends the conversation to LM Studio, gets a text reply, and immediately synthesizes audio for the selected voice.
• /api/voices returns the Kokoro voices parsed from VOICES.md.
• /api/tts runs Kokoro TTS for arbitrary text (used by the "Speak" buttons).

Local chat history, selected model, voice, and temperature are cached in localStorage so your settings persist between refreshes.

Tools and inspiration

• Antigravity - AI coding assistant
• LM Studio - Local LLM server
• Kokoro - High-quality text-to-speech
• LM TTS (LM Studio + Kokoro) by AdmiralApple - Project combining LM Studio and Kokoro