[HuggingFace Demo] [ModelScope Demo]
[Paper] [Demos]

IndexTTS is a GPT-style text-to-speech (TTS) model mainly based on XTTS and Tortoise. It is capable of correcting the pronunciation of Chinese characters using pinyin and controlling pauses at any position through punctuation marks. We enhanced multiple modules of the system, including the improvement of speaker condition feature representation, and the integration of BigVGAN2 to optimize audio quality. Trained on tens of thousands of hours of data, our system achieves state-of-the-art performance, outperforming current popular TTS systems such as XTTS, CosyVoice2, Fish-Speech, and F5-TTS.
Experience IndexTTS: Please contact xuanwu@bilibili.com for more detailed information.

QQ群：553460296
Discord：https://discord.gg/uT32E7KDmy
简历：indexspeech@bilibili.com
欢迎大家来交流讨论！

• 2025/03/25 🔥🔥 We release the model parameters and inference code.
• 2025/02/12 🔥 We submitted our paper on arXiv, and released our demos and test sets.

The overview of IndexTTS is shown as follows.

The main improvements and contributions are summarized as follows:

• In Chinese scenarios, we have introduced a character-pinyin hybrid modeling approach. This allows for quick correction of mispronounced characters.
• IndexTTS incorporate a conformer conditioning encoder and a BigVGAN2-based speechcode decoder. This improves training stability, voice timbre similarity, and sound quality.
• We release all test sets here, including those for polysyllabic words, subjective and objective test sets.

HuggingFace	ModelScope
😁IndexTTS	IndexTTS

Word Error Rate (WER) Results for IndexTTS and Baseline Models on the seed-test

WER	test_zh	test_en	test_hard
Human	1.26	2.14	-
SeedTTS	1.002	1.945	6.243
CosyVoice 2	1.45	2.57	6.83
F5TTS	1.56	1.83	8.67
FireRedTTS	1.51	3.82	17.45
MaskGCT	2.27	2.62	10.27
Spark-TTS	1.2	1.98	-
MegaTTS 3	1.36	1.82	-
IndexTTS	0.937	1.936	6.831

Word Error Rate (WER) Results for IndexTTS and Baseline Models on the other opensource test

Model	aishell1_test	commonvoice_20_test_zh	commonvoice_20_test_en	librispeech_test_clean	avg
Human	2.0	9.5	10.0	2.4	5.1
CosyVoice 2	1.8	9.1	7.3	4.9	5.9
F5TTS	3.9	11.7	5.4	7.8	8.2
Fishspeech	2.4	11.4	8.8	8.0	8.3
FireRedTTS	2.2	11.0	16.3	5.7	7.7
XTTS	3.0	11.4	7.1	3.5	6.0
IndexTTS	1.3	7.0	5.3	2.1	3.7

Speaker Similarity (SS) Results for IndexTTS and Baseline Models

Model	aishell1_test	commonvoice_20_test_zh	commonvoice_20_test_en	librispeech_test_clean	avg
Human	0.846	0.809	0.820	0.858	0.836
CosyVoice 2	0.796	0.743	0.742	0.837	0.788
F5TTS	0.743	0.747	0.746	0.828	0.779
Fishspeech	0.488	0.552	0.622	0.701	0.612
FireRedTTS	0.579	0.593	0.587	0.698	0.631
XTTS	0.573	0.586	0.648	0.761	0.663
IndexTTS	0.744	0.742	0.758	0.823	0.776

MOS Scores for Zero-Shot Cloned Voice

Model	Prosody	Timbre	Quality	AVG
CosyVoice 2	3.67	4.05	3.73	3.81
F5TTS	3.56	3.88	3.56	3.66
Fishspeech	3.40	3.63	3.69	3.57
FireRedTTS	3.79	3.72	3.60	3.70
XTTS	3.23	2.99	3.10	3.11
IndexTTS	3.79	4.20	4.05	4.01

1. Download this repository:

git clone https://github.com/index-tts/index-tts.git

2. Install dependencies:

conda create -n index-tts python=3.10
conda activate index-tts
pip install -r requirements.txt
apt-get install ffmpeg

3. Download models:

Download by huggingface-cli:

huggingface-cli download IndexTeam/Index-TTS \
  bigvgan_discriminator.pth bigvgan_generator.pth bpe.model dvae.pth gpt.pth unigram_12000.vocab \
  --local-dir checkpoints

Recommended for China users. 如果下载速度慢，可以使用镜像：

export HF_ENDPOINT="https://hf-mirror.com"

Or by wget:

wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/bigvgan_discriminator.pth -P checkpoints
wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/bigvgan_generator.pth -P checkpoints
wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/bpe.model -P checkpoints
wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/dvae.pth -P checkpoints
wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/gpt.pth -P checkpoints
wget https://huggingface.co/IndexTeam/Index-TTS/resolve/main/unigram_12000.vocab -P checkpoints

4. Run test script:

# Please put your prompt audio in 'test_data' and rename it to 'input.wav'
PYTHONPATH=. python indextts/infer.py

5. Use as command line tool:

# Make sure pytorch has been installed before running this command
pip install -e .
indextts "大家好，我现在正在bilibili 体验 ai 科技，说实话，来之前我绝对想不到！AI技术已经发展到这样匪夷所思的地步了！" \
  --voice reference_voice.wav \
  --model_dir checkpoints \
  --config checkpoints/config.yaml \
  --output output.wav

Use --help to see more options.

indextts --help

pip install -e ".[webui]"
python webui.py

Open your browser and visit http://127.0.0.1:7860 to see the demo.

On Windows, you may encounter an error when installing pynini: ERROR: Failed building wheel for pynini

In this case, please install pynini via conda:

# after conda activate index-tts
conda install -c conda-forge pynini==2.1.5
pip install WeTextProcessing==1.0.3
pip install -e ".[webui]"

from indextts.infer import IndexTTS
tts = IndexTTS(model_dir="checkpoints",cfg_path="checkpoints/config.yaml")
voice="reference_voice.wav"
text="大家好，我现在正在bilibili 体验 ai 科技，说实话，来之前我绝对想不到！AI技术已经发展到这样匪夷所思的地步了！比如说，现在正在说话的其实是B站为我现场复刻的数字分身，简直就是平行宇宙的另一个我了。如果大家也想体验更多深入的AIGC功能，可以访问 bilibili studio，相信我，你们也会吃惊的。"
tts.infer(voice, text, output_path)

1. tortoise-tts
2. XTTSv2
3. BigVGAN
4. wenet
5. icefall

🌟 If you find our work helpful, please leave us a star and cite our paper.

@article{deng2025indextts,
  title={IndexTTS: An Industrial-Level Controllable and Efficient Zero-Shot Text-To-Speech System},
  author={Wei Deng, Siyi Zhou, Jingchen Shu, Jinchao Wang, Lu Wang},
  journal={arXiv preprint arXiv:2502.05512},
  year={2025}
}