KittenTTS 2 Web UI View on GitHub

Local text to speech that runs on your own GPU.

A small web interface for the KittenTTS 2 speech model. Type a line, pick one of 47 voices, press Generate. Nothing leaves your computer, and there is no account or API key.

Quick start Listen to samples

Unofficial project. Not affiliated with or endorsed by KittenML or Stellon Labs.

The KittenTTS 2 web UI in its ready state: a text box, a voice list, style presets and a Generate button, with a badge showing the model running on an NVIDIA GeForce RTX 3060.
The web UI, ready on an RTX 3060.

What it is

A local server and a single page. It loads the model once, then turns text into speech on your machine.

Offline

After the first model download, it runs without an internet connection. Clips are saved as .wav files in a local folder.

Uses your GPU

Runs on an NVIDIA card with CUDA, using about 4.7 GB of video memory. It falls back to the CPU, slowly.

47 voices

English voices with different characters, plus voices for nine other languages. Each clip can be played, downloaded or deleted from the history list.

Samples

Real output from the app on an RTX 3060. Each clip was generated once and is shown as it came out, apart from trimmed silence and MP3 compression.

Bella

Neutral
“Welcome to KittenTTS 2, a local text to speech studio that runs entirely on your own machine.”
Stable preset

Marcus

Smooth radio
“No cloud, no API keys, no data leaving your computer. Everything runs offline on your GPU.”
Stable preset

Saoirse

Joyful
“Oh, this is the best news I've heard all year!”
Stable preset, no tags

Elliot

Grieving
“I keep thinking about the last time we talked.”
Stable preset, no tags

Victor

Controlled fury
“You lied to me. You looked me in the eye and lied.”
Stable preset, no tags

The mood in these samples comes from the voice, not from markup. Each voice is built from a reference recording made in that mood. The model also accepts emotion tags, vocal events and emphasis markers, but they are a beta feature of the model and in our tests they made little or no audible difference, so none are used here.

Quick start

Setup has two steps: get the code, then run setup.bat once. After that you only need start.bat.

  1. Before you begin, install Python 3.12 (tick Add python.exe to PATH) and, optionally, Git. If you have an NVIDIA GPU, make sure the driver is 570 or newer (run nvidia-smi to check).
  2. Get the code. Open a Command Prompt in a folder on a drive with about 12 GB free and run: 
    git clone https://github.com/UsamaKhanGT4/kitten-tts-webui
cd kitten-tts-webui
    No Git? On the GitHub page click the green Code button, then Download ZIP, and unzip it.
  3. Run setup once. In that folder, run setup.bat (or double-click it). It takes 5 to 15 minutes and downloads a few GB. If it prints a GPU warning, update your NVIDIA driver and run it again.
  4. Every time after that, run start.bat. Your browser opens http://127.0.0.1:7860/. The first start downloads about 2.1 GB of model files, so give it a few minutes.

Requirements

  • Windows 10 or 11. The setup and start scripts are Windows-only.
  • An NVIDIA graphics card with 6 GB of video memory or more is recommended, with driver 570 or newer. Without one it runs on the CPU, slowly.
  • Python 3.12 and about 12 GB of free disk space.