what is voice-builder-for-supertonic-3 fr?

This project is a ready to run Google Colab notebook that trains a custom voice style for Supertonic 3, an open weight text to speech model made by Supertone. It is an unofficial, community built tool, not an official Supertone product, and it is not affiliated with the company. You do not need coding experience or your own GPU, since it runs in Google Colab with a free GPU runtime. The idea is simple: you upload a short recording of a voice, ideally your own or one you have clear permission to use, and the notebook extracts a speaker identity from that recording using an open source optimizer. It then trains a custom voice style file that plugs directly into Supertonic 3, so you can preview it, download it, and use it anywhere Supertonic 3 runs, offline and without ongoing costs. The process happens in eight steps: checking the GPU and installing dependencies, downloading model files, uploading your reference audio, configuring settings, preprocessing the audio, training, testing the new voice, and downloading the trained file. The README is honest about how much quality depends on the input recording. It recommends a quiet room, a phone microphone six to twelve inches away, one to three minutes of natural varied speech, and a single speaker with no background noise or music. It also explains a step where the audio is converted to 16kHz before training, which is required because the speaker identity model needs that format, even though the final generated speech still comes out at much higher quality. The project also states a clear limitation: this method only trains the tone and identity of the voice, not its rhythm and pacing, which come from a starting stock voice, so results are a close likeness rather than a perfect clone. Because Supertonic 3 is released under a license with responsible use rules, the notebook requires users to only clone voices they own or have permission to use, never to impersonate or deceive anyone, never to target minors, and to disclose when generated audio is shared publicly. Sample voice clips in several languages are bundled with the notebook so people can hear results before training their own.