purr

A CLI wrapper for KittenTTS — lightweight ONNX-based text-to-speech on CPU, with model management and optional audio playback.

Installation

Note: Linux only. Requires Python ≥ 3.8 and uv.

Recommended: lean install (no torch/CUDA)

The default install pulls torch and NVIDIA CUDA packages (several GB) as an unnecessary side-effect of kittentts → misaki[en] → spacy-curated-transformers. Use the provided script to bypass this:

git clone https://github.com/newptcai/purr
cd purr
bash install.sh

install.sh installs purr as a uv tool, so it is available on your PATH immediately — no virtual environment activation needed.

Simple install (includes torch/CUDA bloat)

git clone https://github.com/newptcai/purr
cd purr
uv tool install .

Quick Start

📖 Check out repy — a terminal epub reader that uses purr as its default TTS engine.

# List available models
purr model list

# Download the nano model (~50 MB)
purr model install nano

# Synthesize speech to a file
purr speak "Hello, world." --output hello.wav

# Play audio immediately (saves to /tmp/purr-<timestamp>.wav)
purr speak "Hello, world." --play

# Pipe text from stdin
echo "Testing stdin input" | purr speak --play --voice Luna

# Write WAV to stdout for piping into other tools
purr speak "Hello, world." --stdout > hello.wav
purr speak "Hello, world." --stdout | aplay -

# List voices for the active model
purr voices

Commands

purr model

Command	Description
purr model list	Show all models with installation status
purr model install <alias>	Download and install a model
purr model remove <alias>	Remove an installed model

Available model aliases: mini, micro, nano, nano-int8

purr speak

purr speak [TEXT] [OPTIONS]

Option	Short	Default	Description
--model	-m	best installed	Model alias
--voice	-V	Jasper	Voice name
--speed	-s	1.0	Speed multiplier
--output	-o	/tmp/purr-<ts>.wav	Output .wav file
--play / --no-play	-p	off	Play audio after synthesis
--stdout		off	Write WAV audio to stdout (for piping)
--clean / --no-clean		on	Text preprocessing
--quiet	-q	off	Suppress informational output

If TEXT is omitted, input is read from stdin.

purr voices

purr voices [--model nano]

Lists available voices for the specified model.

Model Cache

Models are stored in ~/.cache/kitten-cli/models/ by default. Respects $XDG_CACHE_HOME if set.

License

MIT

Integration

purr is the default TTS tool in repy, a terminal epub reader written in Rust.