GitHub radar
IndexTTS 2.5 Voice cloning with a single audio file.
The Index Team from Bilibili has launched IndexTTS-2.5,an open source TTS model that can clone any voice with one reference sample and support 8 emotions.
IndexTTS-2.5 is an open source TTS model created by the Index Team of Bilibili which can generate a voice with a single reference sample to fit the tone and color of the voice. The model supports 8 kinds of emotions(e.g., happy, sad, angry) controlled by prompts, speech rate control and phonetic accent control. The supported languages include Chinese, English, Japanese, Spanish and Arabic. It doesn’t support Russian. The model requires Python 3.10–3.11 and a GPU with at least 6 GB memory. You could clone the GitHub repo and run uv sync to install.
Why a vibe-coder should care
If you are doing some English videos/podcasts, you have to either generate your own voice or keep the same speaker across episodes. By using IndexTTS-2.5, you can do it with just one reference sample, without relying on any cloud service. You could use it for dubbing, telling stories and building your own TTS system with emotional expressions. Note that IndexTTS-2.5 doesn’t support Russian. If you want to synthesize some Russian voices, you could try other models.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install IndexTTS-2.5 from Index Team (Bilibili): https://huggingface.co/IndexTeam/IndexTTS-2.5 — follow the GitHub README, download the weights, and synthesize 'Hello, this is a test of voice cloning' with zero-shot voice cloning. Ask me if you need a reference audio file.
Requires an NVIDIA GPU with 6 GB of VRAM or more — a MacBook without a discrete NVIDIA GPU won't work.
Open on Hugging Face▌ More finds