Kitten TTS CLI

A simple command-line interface for KittenTTS - a lightweight text-to-speech library.

Installation

Local development

uv sync
uv run kittentts "Your text here"

Global installation

Install as a system-wide tool (without GPU dependencies):

uv tool install --overrides skip-gpu.txt .
kittentts "Your text here"

Usage

Options

Option	Default	Description
text		Text to convert (reads from stdin if not provided)
--model	mini	Model to use (mini, micro, nano, nano-int8)
--voice	Bruno	Voice to use
--speed	1.0	Speech speed multiplier
--output	audio.wav	Output file path (- for stdout)
--mp3		Output as MP3 (replaces .wav with .mp3 or appends .mp3)
--list-voices		List available voices and exit
--normalize		Convert numbers, currencies, abbreviations, etc. to readable text

Available Models

Model	Parameters	Size
mini	80M	80 MB
micro	40M	41 MB
nano	15M	56 MB
nano-int8	15M	25 MB

Voices

Voice names vary by model. Use --list-voices to see available voices for the selected model.

Examples

# Basic usage
uv run kittentts "Hello, world!"

# Use a different voice (use --list-voices to see options)
uv run kittentts --voice Kiki "Hello, world!"

# Use a smaller/faster model
uv run kittentts --model nano "Hello, world!"

# Adjust speed (faster)
uv run kittentts --speed 1.5 "Hello, world!"

# Read from stdin
echo "Hello from stdin" | uv run kittentts
cat input.txt | uv run kittentts

# Pipe audio to stdout
uv run kittentts --output - "Hello" | play -

# Both stdin and stdout
echo "hey" | uv run kittentts --output - | aplay

# Output as MP3
uv run kittentts --mp3 "Hello, world!"  # Creates audio.mp3
uv run kittentts --mp3 --output custom "Hello"  # Creates custom.mp3
uv run kittentts --mp3 --output custom.wav "Hello"  # Creates custom.mp3

# List available voices
uv run kittentts --list-voices

# Convert numbers, currencies, abbreviations, etc. to readable text
uv run kittentts --normalize "I have $100 and 2 cats"

Requirements

• Python 3.12+
• KittenTTS
• soundfile
• ffmpeg (only required for --mp3 option)

Notes

Models are cached in ~/.cache/huggingface/hub/ after first download. Subsequent runs use the cached version.

License

MIT License - see LICENSE file for details.