fawzan09/-supertonic-3-tts-audiobook-studio

This project is a no-code Google Colab notebook that wraps Supertonic 3, an open-weight text-to-speech model from Supertone Inc, into a full audio production studio you operate entirely through point-and-click forms, with no code editing needed. It is an independent, unofficial notebook and is not made or endorsed by Supertone Inc. The notebook is organized into fifteen numbered cells, each handling one task. After a one time setup cell that installs everything and downloads the roughly 400 megabyte model, you get simple text-to-speech generation, a way to preview every built-in voice, an experimental tool that blends two voices together, and a text cleanup step that accepts PDF or plain text files and strips out page numbers, repeated headers, and broken line wraps. The core feature is an audiobook generator that turns a text, EPUB, or PDF file into fully narrated audio split by chapter, with subtitles, and a separate cell can package those chapters into a real M4B audiobook file with chapter markers and cover art, playable in apps like Apple Books. Other cells handle scripted multi-voice dialogue scenes, batch converting a folder of text files, loudness normalizing the audio to broadcast standards, mixing in intro and outro music, and even running Supertonic as a live API server with an optional public web address. Getting started only requires a free Google account and no GPU, since the underlying model is designed to run on a CPU. You simply open the notebook in Colab, run the setup cell, then fill in each form and press run. Output files are saved automatically and offered as downloads. For a closer voice match than the built in options, the author points to a separate companion tool that trains a custom voice from a longer reference recording, sold as a one time purchase, though it is optional and not required to use this notebook. The notebook itself is built on Supertonic's reference code, released under the MIT License, while the model weights use a separate OpenRAIL-M license that allows commercial use but restricts harmful use and non-consensual voice impersonation, with an attribution requirement.