Skip to the content
Know your tech before you buy it. Makers: get listed from $49/mo

IndexTTS

github.com

  • Free plan
  • Windows
  • Linux
  • Web
  • API
  • Self-hosted
8.0 Club score Visit github.com

What IndexTTS is

IndexTTS is a text-to-speech system that creates speech by cloning a voice from one reference audio clip. IndexTTS-2.5 supports Chinese, English, Japanese, Spanish, and Arabic, with WAV as its listed export format. You can guide emotion with a reference recording, an emotion vector, or text input. Speed adjustment uses duration_factor values from 0.5 to 2.0, where 1.0 is normal. Pronunciation controls include Chinese Pinyin, English CMU phonemes, and Japanese Kana; for IndexTTS-2, Pinyin control applies only to cases supported in the project's vocabulary file.

The project provides a browser-based WebUI, a Python API, and a vLLM recipe for production serving. Installation guidance covers Windows and Linux, and model downloads are available through Hugging Face or ModelScope. Setup calls for Git and uv; CUDA Toolkit 12.8 or newer is recommended for CUDA installation errors. DeepSpeed is optional, with results varying by hardware, drivers, and operating system. IndexTTS is free and open source under the bilibili Model Use License Agreement; users are asked to read the disclaimer before use. The project directs commercial users to contact its team.

Who it suits

IndexTTS suits developers or teams seeking voice cloning and speech generation through a local WebUI, Python API, or production serving setup. Its pronunciation controls and five listed languages may suit projects that need those specific options.

Good to know, and what to watch

  • Clones a voice from one reference audio clip
  • Supports five languages in IndexTTS-2.5
  • Offers emotion and speaking-speed controls
  • Provides a WebUI, Python API, and vLLM recipe
  • Pinyin control covers only listed supported cases
  • Setup calls for Git and uv
  • DeepSpeed may be difficult to install on Windows

IndexTTS

Free

Open source

Getting started

  1. Open the IndexTTS GitHub repository
  2. Install Git and uv
  3. Follow the Windows or Linux installation guidance
  4. Download models from Hugging Face or ModelScope
  5. Use the browser WebUI or Python API
  6. Configure CUDA Toolkit 12.8 or newer when required

Limits to check

The project lists IndexTTS at no cost and provides downloadable model and inference code rather than a hosted plan. Pinyin control for IndexTTS-2 works only for supported cases in its vocabulary file.

The Club's review

IndexTTS is a self-hosted text-to-speech system that clones a voice from a reference clip. It suits developers and technically confident creators who want control over speech generation. Its range of voice and delivery controls is compelling, but setup and license terms make it a less straightforward choice for commercial or hosted workflows.

Overview

IndexTTS turns text into speech in a cloned voice, with IndexTTS-2.5 supporting Chinese, English, Japanese, Spanish, and Arabic. The combination of voice cloning and adjustable delivery makes it more useful for tailored narration than a basic speech generator. Output is available in WAV.

The project provides downloadable models and inference code at no charge. That favors people who can run and maintain their own installation; there is no hosted plan or usage quota to rely on. The project is released under the bilibili Model Use License Agreement, and commercial users are directed to contact the project about commercial use and cooperation.

Key features

Voice, emotion, and pace

A single reference audio clip is the basis for voice cloning. Emotion can be guided with an emotional reference recording, an emotion vector, or text input, giving creators multiple ways to shape a performance. The speed control, duration_factor, ranges from 0.5 to 2.0, with 1.0 as normal speed. These options help when narration needs a specific voice or delivery, though they require configuring an inference workflow rather than selecting a hosted preset.

Pronunciation control

IndexTTS-2.5 supports Chinese Pinyin, English CMU phonemes, and Japanese Kana. That can help with names and terms that ordinary text input might pronounce poorly. Pinyin control for IndexTTS-2 works only for supported cases in the project's vocabulary file, so it is not a general fix for every Chinese pronunciation.

Ways to run it

The project documents a browser-based WebUI, a Python API, and production serving through a vLLM recipe. This gives individual users and developers different routes into inference, including a path toward production serving. Setup calls for Git and uv; CUDA installations on Windows or Linux may require CUDA Toolkit 12.8 or newer when a CUDA installation error occurs. Model downloads are supported through Hugging Face or ModelScope.

DeepSpeed is optional and may speed up inference on some systems, but results depend on hardware, drivers, and operating system. The README warns that installing DeepSpeed can be difficult on Windows, which makes that platform a less predictable choice for users who need the acceleration.

Pricing

IndexTTS is free and open source. The IndexTTS plan costs nothing and provides downloadable models and inference code; there is no hosted plan or published price. The free option suits users able to supply their own computing environment, but it does not remove the work of installation and running inference. Review the bilibili Model Use License Agreement and disclaimer before use. For commercial usage and cooperation, the project directs users to contact [email protected].

Platforms

IndexTTS supports API access, self-hosting, Linux, Windows, and a browser-based WebUI. Windows and Linux installation guidance is provided, though DeepSpeed may be difficult to install on Windows. The maintainers identify the GitHub repository as their only official channel and do not guarantee the security, accuracy, or timeliness of other websites or services.

Who it's for

IndexTTS is best suited to developers and technically confident creators who want to run voice cloning locally, call inference through Python, or serve it through a production workflow. It is a strong fit when emotion, pace, or pronunciation control matters and the user can manage models and hardware. Those seeking a managed speech service, a simple setup, or clear commercial licensing terms should consider other options.

Pros and cons

  • Pros: Voice cloning starts from a single reference clip, and emotion, speed, and pronunciation controls support more tailored output.
  • Pros: A WebUI, Python API, and vLLM serving recipe provide options for individual use and deployment.
  • Pros: The downloadable models and inference code are free, with no stated hosted usage quota.
  • Cons: Users must handle installation, model downloads, and their own inference environment.
  • Cons: DeepSpeed setup may be difficult on Windows, and acceleration depends on the system.
  • Cons: Pinyin control is limited to supported cases, while commercial users need to contact the project about use and cooperation.

Alternatives

Fish Audio is a freemium alternative with API, self-hosted, browser, Linux, macOS, and Windows support; its free tier includes 8,000 monthly credits, up to seven minutes of generation, and a 500-character limit per generation.

Kits AI may suit users focused on generative vocals: its free plan includes voice designer/blender and Kits generative vocals, with 15 conversion minutes but no voice slots or download minutes.

Qwen3-TTS is another free, self-hostable option, with 0.6B and 1.7B models and support for 10 languages under the Apache-2.0 license.

Sprag is a paid API and self-hosted option for stock-voice text-to-speech, with its Chorus plan priced at $0.70 per hour of generated speech.

Voicebox is another option for readers comparing speech software across desktop, mobile, API, and self-hosted platforms.

VoxCPM is a free, open-source alternative with API, Linux, macOS, self-hosted, and web support.

Fish Speech offers a free tier with 8,000 monthly credits, up to seven minutes of generation, up to 500 characters per generation, three public voice slots, and standard generation speed.

VoiceStudio is another free option with API, Linux, macOS, self-hosted, web, and Windows support.

For more options, browse voice cloning software or text-to-speech software.

Verdict

Choose IndexTTS if you can manage a local inference setup and want voice cloning with meaningful control over emotion, speed, and pronunciation. The free downloadable models and flexible deployment paths are its clearest strengths. Look elsewhere if you need a hosted service or straightforward commercial terms; the license and setup requirements call for care before adoption.

The verdict

IndexTTS combines voice cloning with controls for emotion, speed, and pronunciation, and offers both browser and API access. Check its license and setup requirements, and confirm that its language and pronunciation support fit your use case.

Questions people ask

Is IndexTTS free?

Yes. The listed IndexTTS plan is free, and the project is open source.

Which languages does IndexTTS-2.5 support?

Chinese, English, Japanese, Spanish, and Arabic.

What output format is available?

The listed export format is WAV.

Can I use an API?

Yes. The repository provides a Python API and documents production serving through a vLLM recipe.

What license applies?

The project uses the bilibili Model Use License Agreement and asks users to read its disclaimer. Commercial users are directed to contact [email protected].

What the maker publishes

Acceleration
Optional DeepSpeed support may speed up inference on some systems, but the repository says results depend on hardware, drivers, and operating system. source
API access
Yes source
Commercial use
For commercial usage and cooperation, the project directs users to contact [email protected]. source
Deployment
The project documents a local WebUI, a Python API, and production serving through a vLLM recipe. source
Emotion control
Speech emotion can be controlled with an emotional reference recording, an emotion vector, or text. source
Export formats
WAV source
Hardware
The README recommends NVIDIA CUDA Toolkit 12.8 or newer on Linux or Windows when a CUDA installation error occurs. source
Installation platforms
The README gives installation guidance for Windows and Linux and notes that DeepSpeed may be difficult to install on Windows. source
Interfaces
The repository provides a browser-based WebUI and a Python API for inference. source
Languages
IndexTTS-2.5 supports Chinese, English, Japanese, Spanish, and Arabic. source
License
The project says it is released under the bilibili Model Use License Agreement and asks users to read its disclaimer before use. source
Model downloads
The README provides model download instructions using Hugging Face or ModelScope. source
Notable pronunciation limit
For IndexTTS-2, Pinyin control works only for supported Chinese Pinyin cases listed in the project's vocabulary file. source
Official channel
The maintainers identify the GitHub repository as the only official channel maintained by the core team and say other sites or services are not official. source
Official channel and security
The maintainers say the GitHub repository is their only official channel and that they cannot guarantee the security, accuracy, or timeliness of other websites or services. source
Platforms
web,windows,linux,api,self_hosted source
Product
IndexTTS is a zero-shot text-to-speech system that clones a voice from a single reference audio clip. source
Pronunciation
IndexTTS-2.5 supports pronunciation control using Chinese Pinyin, English CMU phonemes, and Japanese Kana. source
Purpose
IndexTTS is a zero-shot text-to-speech system that clones a voice from a single reference audio clip. source
Requirements
The setup instructions call for Git, uv, and, for Linux or Windows CUDA installations, CUDA Toolkit 12.8 or newer. source
Speaking speed
IndexTTS-2.5 supports duration_factor values from 0.5 to 2.0, with 1.0 as normal speed. source
Speed control
IndexTTS-2.5 supports speaking speed adjustment through duration_factor from 0.5 to 2.0. source
Support
The README lists QQ groups, a Discord server, and [email protected] as community contact options. source
Voice and emotion controls
The project describes fine-grained emotion control using emotional reference audio, emotion vectors, or text-based emotion input. source

Checked against the maker's pages on 9 Oct 2026.