How to Use Index TTS: Complete Step-by-Step Guide
Want to know how to use Index TTS without getting lost in technical settings? This guide walks through the complete process from preparing your first script to generating, reviewing, and improving AI speech.
Before You Start
Begin by deciding what you want to create.
Are you making a simple narration? Do you want to clone an authorized voice? Are you producing emotional character dialogue? Do you need multilingual speech? Do you need tighter control over speaking speed?
Knowing the goal makes the rest of the setup easier.
You should also prepare a clean voice recording if your project uses voice cloning. Only use audio you own or have permission to use.
IndexTTS 2 · expressive emotion
Best for emotional delivery, emotion reference audio, and duration control.
IndexTTS 2.5 · 5 languages + speed
Chinese, English, Japanese, Spanish, Arabic with speaking-speed control.
Authorized reference voice
Only upload voices you own or have permission to use.
The 12-Step How to Use Index TTS Workflow
From choosing the model to correcting pronunciation — one small decision at a time.
Choose the Right Model
Select the model that matches your project. Choose IndexTTS 2 when emotional performance, separate emotion and speaker references, character dialogue, or expressive narration matters. Choose IndexTTS 2.5 when you need multilingual speech, speaking-speed control, newer pronunciation options, or faster generation.
Prepare Your Script
Write text for listening rather than silent reading. Use complete sentences, add commas where a short pause sounds natural, use periods to separate ideas, and break very long passages into smaller sections. For your first test, use only one or two sentences.
Prepare Reference Audio
Reference quality has a major effect on voice cloning. Choose a recording containing one speaker, clear speech, stable volume, minimal background noise, and limited room echo. Avoid clips containing music, several speakers, heavy processing, clipping, or strong environmental noise.
Add the Speaker Reference
Upload your authorized reference recording into the speaker or voice field. This input tells the system what vocal identity should guide the generated speech. Do not confuse speaker reference audio with emotional reference audio — speaker reference = who should speak; emotion reference = how the line should feel.
Select the Language
When using IndexTTS 2.5, select the language that matches your text. If you are creating cross-language speech, review pronunciation carefully after generation. Names and mixed-language terminology may require additional attention. Start with one language during your first tests.
Generate a Baseline
Before using advanced features, generate one simple version using a clean speaker reference, short text, default pacing, and minimal additional controls. Then listen to the entire clip: does the speaker sound consistent? Are the words clear? Does the sentence flow naturally?
Add Emotional Reference Audio
If the voice needs more expression, add an emotion reference where supported. Use a clip that clearly represents the desired emotional style. You can keep the original speaker reference while using a different emotional source — identity is maintained while delivery changes.
Adjust Emotion Strength
Do not assume that stronger emotion always sounds better. Try several levels. A tutorial may need almost no emotional exaggeration, a fictional character may need much more, and a serious advertisement might require confidence without sounding theatrical.
Use Structured Emotion Controls
Some workflows provide direct emotion parameters instead of requiring an emotional audio clip. These can represent emotional dimensions such as happiness, anger, sadness, fear, surprise, or calmness. Start with one clear direction, then experiment with combinations.
Use Text-Based Emotion Guidance
If you do not have suitable emotion audio, a text description can be useful when supported. Try descriptions such as calm and reassuring, excited but controlled, soft and melancholic, or nervous and uncertain. Keep descriptions consistent with the meaning of the sentence.
Adjust Speaking Speed
IndexTTS 2.5 provides control over speaking pace. Start close to normal speed, then make small changes. A tutorial may benefit from slightly slower delivery, while a short promotional video may need faster pacing. Do not jump immediately to extreme values.
Correct Pronunciation Problems
Generate normal text first. If a word sounds wrong, identify the exact problem instead of modifying the entire script. Difficult cases often include names, brand terms, technical vocabulary, abbreviations, polyphonic characters, and foreign words. Use pronunciation guidance only where needed.
Long Text and Voice Cloning Workflows
How to Use Index TTS for Long Text
For long narration, divide the script into sections.
A practical structure is: Introduction, Section 1, Section 2, Section 3, Conclusion.
Generate each section separately while keeping the same reference audio and core settings.
This has several advantages. If one sentence is wrong, you only regenerate that section. Emotional transitions are easier to control. Pronunciation mistakes are easier to identify. Audio files are easier to organize.
Long-form generation should prioritize consistency rather than generating the largest possible block of text at once.
How to Use Index TTS for Voice Cloning
For authorized voice cloning:
Choose clean reference audio. Upload it as the speaker reference. Enter a short test script. Generate a baseline. Check speaker similarity. Try another reference clip if necessary. Add emotional controls only after the base voice is stable.
Do not try to fix poor voice similarity using unrelated emotion or speed settings.
Start with the reference recording.
How to Use Index TTS Locally
Developers who want more control can run the official project locally. The project currently uses uv for dependency management.
The current official installation begins by cloning the repository:
git clone https://github.com/index-tts/index-tts.git
cd index-ttsThen install dependencies with uv:
pip install -U uv
uv sync --all-extrasUnderstanding how to use Index TTS locally requires a suitable Python/GPU environment, so a browser-based service may be simpler if you only want to generate speech.
Download IndexTTS 2.5 and IndexTTS 2
Download IndexTTS 2.5
For how to use Index TTS 2.5 locally, the official repository currently documents:
uv tool install "huggingface-hub"
hf download IndexTeam/IndexTTS-2.5 --local-dir=checkpointsThe model is placed in the checkpoints directory. After the files are available, the default WebUI uses IndexTTS 2.5.
Download IndexTTS 2
For how to use Index TTS 2, the official repository uses a different local checkpoint directory:
hf download IndexTeam/IndexTTS-2 --local-dir=checkpoints_2Keeping the directories separate makes it easier to switch between the models — useful when comparing the expressive behavior of IndexTTS 2 with the newer multilingual and speed-control features of 2.5.
Launch the IndexTTS WebUI
The simplest local answer to how to use Index TTS is often the official WebUI.
For the default IndexTTS 2.5 model:
uv run webui.pyFor IndexTTS 2:
uv run webui.py --version 2 --model_dir ./checkpoints_2Then open http://127.0.0.1:7860. Once the interface loads, how to use Index TTS becomes similar to using an online generator: provide inputs, adjust settings, generate, listen, and refine.
Check GPU Acceleration
If local generation is unexpectedly slow, part of how to use Index TTS is confirming that your hardware is detected correctly.
Run the built-in GPU check:
uv run tools/gpu_check.pyThe repository also documents BF16 inference for IndexTTS 2.5 and FP16 for IndexTTS 2 as lower-VRAM options, with faster inference and a small quality tradeoff. Performance will still depend on your hardware, drivers, operating system, and configuration.
Common How to Use Index TTS Problems
The Voice Does Not Match the Reference
Try a cleaner recording. Reduce background noise. Use one speaker only. Avoid music or strong echo. Reference quality should be checked before changing advanced settings.
The Voice Sounds Too Emotional
Reduce the emotional influence. Try a more subtle reference. Compare different strengths instead of using the maximum value.
A Name Is Pronounced Incorrectly
Check spelling first. Then use pronunciation controls when the language and model support them. Test the difficult word in a short sentence before generating an entire section.
The Audio Is Too Fast
Use the available speaking-speed controls in IndexTTS 2.5. Make gradual adjustments.
Long Narration Sounds Inconsistent
Divide the script into smaller sections. Keep reference audio and core generation settings consistent across the project.
Local Generation Is Slow
Check whether GPU acceleration is working correctly. Your hardware, model version, precision settings, and environment can all affect performance.
Best Practices
A reliable workflow is simple: good reference → short test → baseline → emotion → speed → pronunciation → final generation.
Change one variable at a time.
If you change reference audio, emotion, speed, and text simultaneously, you will not know which adjustment improved or damaged the result.
The most effective users treat AI voice generation as an iterative production process.
Change one variable at a time
The most effective way to improve your workflow — never change several settings at once.
Follow a test sequence
Good reference → simple text → baseline check → emotion → speed → pronunciation.
Know which change mattered
Isolating variables makes troubleshooting easier because each change is traceable.
Experiment deliberately
Controlled experimentation produces more useful results than random adjustments.
How to Use Index TTS Responsibly
Learning how to use Index TTS also means understanding responsible voice use.
Owned or authorized voices
Only upload a person's voice when you own the recording or have appropriate authorization to use it for your intended purpose.
No deceptive impersonation
Do not use cloned voices to impersonate people deceptively or misrepresent who created a piece of speech.
Support legitimate creativity
AI voice generation is most useful when it supports legitimate creativity, accessibility, experimentation, and production rather than deception.
How to Use Index TTS FAQ
Start with a short script and clean reference audio. Generate a simple baseline before changing emotion, speed, or pronunciation settings.
Start Using Index TTS
The easiest way to understand how to use Index TTS is to begin with one simple generation.
Choose the right model. Prepare clean text. Add authorized reference audio. Generate a short sample. Listen carefully. Then introduce more advanced controls only when your project needs them.
Following this process makes voice cloning, emotional speech, multilingual generation, and longer narration much easier to manage.
References: GitHub · index-tts/index-tts README · Official IndexTTS page · GitHub · index-tts/index-tts