KittenTTS Setup

Back to the TTS Guide

KittenTTS runs a neural TTS model directly on your machine. No cloud, no API keys, no data leaving your computer — and the voice quality is genuinely good. Setup takes about 5 minutes. English only for now.

Hardware Requirements

KittenTTS uses PyTorch for neural network inference on your CPU. This means it needs real computing power:

Hardware	Experience
8+ cores, modern CPU (Ryzen 7, i7, Xeon)	Excellent. Speech generates quickly, minimal impact on other tasks
4-6 cores (Ryzen 5, i5)	Good. Noticeable CPU usage during generation but perfectly usable
2 cores / older CPU	Slow. Generation may take several seconds per utterance. Consider Google Cloud instead

CPU usage is temporary — KittenTTS only uses CPU while actively generating speech (typically 1-2 seconds per utterance). Between utterances, CPU usage drops to near zero.

Thread Management

By default, KittenTTS uses all available CPU cores for maximum speed. If you’re also running a chess engine (like Stockfish), you may want to limit the threads KittenTTS uses.

In Settings > Sound > KittenTTS CPU Threads, set a value to cap thread usage. Set to 0 for automatic (use all cores). A good starting point for shared use with a chess engine is half your core count.

Tip: KittenTTS and Stockfish both want CPU cores. If you’re analyzing at full depth while KittenTTS generates speech, both will compete for CPU time. On a machine with 8+ cores, you’ll rarely notice. On 4 cores, give KittenTTS 2 threads and leave the rest for the engine.

Step 1: Install Dependencies

KittenTTS requires Python 3.10+ and a few Python packages. Choose one of these methods:

Option A: In-App Setup Wizard (Recommended)

1. Open En Parlant~ and go to Settings > Sound
2. Set TTS Provider to KittenTTS (English Only)
3. If dependencies are missing, a yellow “Setup Guide” alert appears
4. Click the alert to open the setup wizard
5. The wizard walks you through each step with “Fix” buttons for automatic installation

Option B: Terminal Setup Script

cd /path/to/en-parlant
./scripts/setup-tts.sh --kittentts

This creates a Python virtual environment and installs the required packages (kittentts, flask, soundfile, numpy). The nano model (~25MB) downloads from HuggingFace on first run.

Option C: Manual Setup

cd /path/to/en-parlant/scripts
python3 -m venv .venv
.venv/bin/pip install kittentts flask soundfile numpy

Step 2: Configure En Parlant~

1. Open En Parlant~ and go to Settings (gear icon) > Sound tab
2. Set TTS Provider to KittenTTS (English Only)
3. The server starts automatically when you select this provider
4. Wait a few seconds for the model to load (first run downloads from HuggingFace)
5. Choose a voice — there are 8 options (4 male, 4 female)
6. Click the Test button next to the voice selector

You should hear a chess move spoken aloud with natural, expressive AI speech.

Troubleshooting

• “Server not responding” error? The KittenTTS server may not have started. Check the setup wizard in Settings — it shows dependency status and can auto-fix missing packages.
• First run is slow? The ~25MB nano model downloads from HuggingFace on first use. This is a one-time download. Subsequent starts take 2-5 seconds.
• High CPU usage? This is normal during speech generation. Reduce the thread count in Settings if it’s impacting other tasks.
• No sound? Make sure Python 3.10+ is installed and the virtual environment was created successfully. The setup wizard can diagnose this.

Voice Guide

KittenTTS offers 8 voices numbered 2-5, each in male and female variants. All voices are English with slightly different tonal qualities. Try a few to find the one you like best.

Language Note

KittenTTS currently supports English only. For other languages, use ElevenLabs or Google Cloud. The TTS language setting is ignored when using KittenTTS — chess terms are always spoken in English.