what is colab-tts-api fr?

This project lets you run a text-to-speech server that can clone voices. You write some text, provide a short audio sample of a voice, and the server produces spoken audio matching that voice. It supports three different AI models for this: Chatterbox (a faster, smaller model), VibeVoice (a larger model focused on expressive speech), and Dia (an English-only model that needs a written transcript of your reference audio). Each model runs in its own isolated environment, so you can switch between them without reinstalling anything. The server is designed to run on Google Colab, a local Linux machine, or any computer with a compatible Nvidia GPU. You launch it by running a single script per model. On first run, it downloads the model and its dependencies, which takes about 30 to 60 seconds. After that, it prints a public web URL you can use to send requests to the server from anywhere. The project uses Cloudflare to create this public tunnel, so you do not need to configure port forwarding yourself. The API offers two ways to interact with it. The first is a custom endpoint where you send text and optionally an audio file of a voice to clone. You can also upload voice clips once, give them names, and reuse them later. The second option is an OpenAI-compatible endpoint, meaning it works with existing software built for OpenAI's text-to-speech API. You send a JSON request with your text and voice choice, and it returns audio in formats like WAV, MP3, or FLAC. Configuration is handled through optional environment variables. You can adjust things like the output port, which CUDA version to use, and model-specific settings such as inference steps or temperature. The project also includes a simple graphical interface and a command-line tool for generating audio from a text file. Developers can add new models by following a few steps: creating a new backend class in the server code, copying a launcher script, and registering the new model in the dispatcher.