wtf is pocket-tts?

Pocket TTS is a text to speech tool from Kyutai that turns written text into spoken audio using only a computer's CPU. It skips the need for a GPU or a web API, so you install it with pip and start generating audio with a single function call. The project supports Python versions 3.10 through 3.14 and needs PyTorch 2.5 or newer, but not the GPU version of PyTorch. The core model is small, about 100 million parameters, which keeps it light enough to run fast even on modest hardware. On a MacBook Air M4 CPU it generates audio about 6 times faster than real time, using only 2 CPU cores, and the first chunk of audio appears in around 200 milliseconds. It streams audio as it generates, so you do not have to wait for the whole clip to finish before playback starts. It can also handle text of any length. You can try Pocket TTS three ways. The website lets you type text, pick a voice, and hear the result with no installation at all. The command line tool offers a generate command for quick one off audio files and a serve command that starts a local web server, keeping the model loaded in memory so repeated requests are much faster. There is also a Python library you can install and call directly in your own code, shown in the README with a short example that loads a voice and produces an audio array you can save as a wav file. A standout feature is voice cloning: you can pass any wav file as a voice reference, or choose from a catalog of premade voices, and the model will speak in that voice. Pocket TTS supports six languages, english, french, german, portuguese, italian, and spanish, with plans to add more later. Some non english languages also offer a larger, higher quality variant that runs slower. There is also a browser based implementation for running client side, plus full documentation and a published tech report and paper for anyone who wants technical depth.