- Developer
- bilibili IndexTTS
- Version covered
- v1.5.0 / IndexTTS-1.5
- Main languages
- Chinese and English
- Checkpoint
- IndexTeam/IndexTTS-1.5
Strengths worth exploring
Correct Chinese readings with pinyin
Mixed character/pinyin modeling enables targeted pronunciation changes. Names, place names, and ambiguous characters often benefit more from such precise input than from changing the whole voice reference.
Official source [1]A focused stability and English update
The release identifies stability and English synthesis as its main improvements. It does not introduce the separate emotion-control workflow of IndexTTS 2.
Official source [1]Clone without speaker-specific training
Speaker conditioning and a BigVGAN2-based decoder support synthesis from reference speech, aiming to preserve the voice while producing a new script.
Official source [2]A narration workflow with explicit correction
Explainers, training scripts, product names, and mixed-language narration need understandable words before they need dramatic acting. Pinyin, punctuation, and the reference recording provide different handles on the result.
Pronunciation correction is a specific input technique, not a general natural-language style prompt. Listen separately for reading accuracy, pauses, and voice similarity.
Official source [1]Keep 1.5 code and weights together
This guide links to v1.5.0 code and the IndexTTS-1.5 checkpoint. Version 2 uses a different inference entry point and emotion parameters. Installing a newer branch and following an older example can obscure which model is actually being evaluated.
Official source [1]Usage and deployment
- The example below targets Linux/NVIDIA. Install Git and Conda, then check the PyTorch/CUDA combination supported by your device.
- Check out v1.5.0, create a Python 3.10 environment, install FFmpeg, PyTorch, and WebUI dependencies, and download IndexTTS-1.5 into checkpoints.
- Launch webui.py and open localhost port 7860. For a Python integration, use indextts.infer.IndexTTS from this release.
git clone --branch v1.5.0 https://github.com/index-tts/index-tts.git
cd index-tts
conda create -n indextts15 python=3.10
conda activate indextts15
conda install -c conda-forge ffmpeg
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu118
pip install -e '.[webui]' --no-build-isolation
pip install 'huggingface_hub[cli]'
huggingface-cli download IndexTeam/IndexTTS-1.5 --local-dir checkpoints
python webui.pyBefore you get started
- The CUDA 11.8 example follows this release's documentation; verify driver and PyTorch compatibility before using it.
- Independent emotion references and eight-dimensional emotion vectors belong to version 2, not 1.5.
Where it fits
Useful for Chinese/English narration, targeted pronunciation correction, or an existing 1.x workflow that needs a reproducible release.
Official resources and sources
DS Speech currently provides Qwen-Audio-3.0-TTS online. Other models in this library are documented for learning and self-deployment.