Pocket TTS is a lightweight text-to-speech engine designed to run efficiently on CPUs without requiring GPUs or external web APIs.
The tool addresses the friction of deploying TTS systems by packaging a compact 100-million-parameter model that generates speech through a simple pip install and function call. It streams audio output, delivering the first chunk in roughly 200 milliseconds and generating speech faster than real-time on modest hardware like a MacBook Air M4, using only 2 CPU cores. The model supports eight languages including English, French, German, Portuguese, Italian, and Spanish, handles arbitrarily long text inputs, and includes voice cloning capabilities.
Pocket TTS suits developers building applications that need on-device speech synthesis without infrastructure overhead. It works well for edge deployments, client-side browser implementations, and scenarios where latency and independence from cloud services matter. The tool requires Python 3.10 or later and PyTorch 2.5 or newer, but does not mandate the GPU version of PyTorch. A command-line interface and Python API both ship with the package, and a web demo is available for trying the tool without installation.
The project maintains active development with recent releases of training code that allow users to train custom models, and it actively solicits community contributions of trained models. The codebase includes comprehensive documentation, a technical report, and an academic paper describing the approach. The tool is available through standard Python package management and can run directly in web browsers for client-side use.
