Description
IndexTTS-2.5 is a sophisticated text-to-speech model developed by IndexTeam and Bilibili. It is designed to clone voices from a single reference audio clip, making it a powerful tool for voice synthesis. The model supports multiple languages, including Chinese, English, Japanese, Spanish, and Arabic, allowing for cross-lingual voice transfer. One of its key features is emotion control, which is disentangled from timbre, providing users with the ability to manipulate the emotional tone of the synthesized speech.
Read More
Compared to its predecessor, IndexTTS-2, this version introduces support for additional languages such as Japanese, Spanish, and Arabic. It also offers faster inference times and improved controllability over Chinese Pinyin, English CMU phonemes, and Japanese Kana. The model operates with a GPT backbone and uses a flow-matching speech-to-mel decoder along with a BigVGAN vocoder, boasting approximately 0.8 billion parameters.
IndexTTS-2.5 outputs a 22.05 kHz waveform and requires Python 3.10–3.11, an NVIDIA GPU, and about 6 GB of VRAM for inference. Users need to download the model weights, and auxiliary models are downloaded into checkpoints on the first run. The model's limitations include the need for segmenting long texts, which affects prosody across segment boundaries, and the requirement for user responsibility in obtaining consent for voice cloning.
The model is licensed under the bilibili Model Use License Agreement, and its repository is available on GitHub. With nearly 15,000 downloads in the last month, IndexTTS-2.5 is a popular choice for developers and researchers interested in advanced text-to-speech capabilities.
- CategoryText to Speech
- Pricing ModelOpen source
- DeploymentCloud
- PlatformsWeb
IndexTTS-2.5 Highlights
Zero-shot voice cloning
Supports Chinese, English, Japanese, Spanish, Arabic
Cross-lingual voice transfer
Emotion control disentangled from timbre
Faster inference times
Improved controllability of phonemes
22.05 kHz waveform output
Autoregressive TTS with GPT backbone
Getting Started with IndexTTS-2.5
Access model: Visit the Hugging Face page
Authenticate: Set up access credentials
Set up environment: Install Python 3.10–3.11 and required libraries
Integrate via API: Use the provided API for integration
Optimise: Adjust parameters for desired output
IndexTTS-2.5's Use Cases
- Voice Cloning
- Multilingual Synthesis
- Emotion Control
- Cross-Lingual Transfer
- Phoneme Control
