Vakya 2.0 - Text-to-Speech

Vakya 2.0 - Text-to-Speech

Turns text in 11 Indian languages into speech, cloning a voice from a short reference clip. For researchers and learners who have permission to use the voice they clone.

Made by YashMachineLearning on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: Vakya 2.0 is a text-to-speech playground built on the IndicF5 design from AI4Bharat. You type text in one of 11 Indian languages, give it a short reference clip, and it speaks the text in that voice. The model has 0.4B parameters and outputs 24 kHz audio.

Who it suits: It suits researchers and students who want to try speech in Indian languages and voice cloning, always with a voice they have permission to use.

Industry | Who uses it | How they use it | Example
Research | Speech researchers | Test voice cloning across Indian languages | Clone a short clip and read a Tamil sentence
Education | Teachers and students | Hear text read aloud in a regional language | Listen to a Hindi lesson passage
Content | Creators with their own voice | Voice a script in a language they do not speak | Narrate a Marathi caption in their own voice
Development | Developers | Evaluate a small TTS model before building on it | Check how speed settings change the result

You can skip it if you need a language outside its list, or want to clone a voice without the speaker's permission.

Where to use it:

• A web browser, on the Space's page at Hugging Face
• Your own computer, on Windows, macOS or Linux

Use case | Who | What they get back
Clone a voice | Researchers | Speech in a voice close to the reference clip
Read text aloud | Learners | Audio of text in a supported language
Tune playback | Developers | Faster or slower speech, from 0.5x to 2.0x

How to set it up:

• Nothing to install if you use the Space online
• Python, if you run a copy on your own computer
• The Python packages in requirements.txt
• The system packages in packages.txt, which Hugging Face installs with apt-get

1. Open the Space's page on Hugging Face to use it online.
2. To run a copy, install huggingface_hub and download the Space into a folder.
3. Move into that folder and install Gradio at the pinned version, then the listed packages.
4. Start the app with Python.

pip install -U huggingface_hub
hf download YashMachineLearning/indicF5 --repo-type space --local-dir indicF5
cd indicF5
pip install gradio==6.2.0
pip install -r requirements.txt
python app.py

How to use it: Work down the page in order. Load the model first, which can take a few minutes on the first run. Then add your reference clip, type your text and generate. Keep the clip under 15 seconds.

Tool | What it does
Load Model | Downloads and prepares the model before first use
Reference audio | A short clip of the voice to clone
Reference text | Optional words spoken in the clip, left blank to be transcribed automatically
Text to generate | The text to speak, in any supported language
Speed | Sets the speech rate from 0.5x to 2.0x
Remove Silences | An experimental option that trims pauses
Generate Speech | Makes the audio

Example: Click Load Model, upload a clip of your own voice under 15 seconds, type a Hindi sentence in the text box and click Generate Speech. You get audio of the sentence in a voice close to your clip.

For newcomers and experts: If you are new, start with your own voice clip, since you clearly have the right to use it. Clear audio helps, and adding the reference text helps the voice match.

If you know the field, the model follows the IndicF5 architecture and the README says it was trained on 1417 hours of speech. It names Whisper, Qwen2.5 3B Instruct, Vocos and BigVGAN among the models the Space lists, though the README does not say what each does. The Space runs on basic CPU hardware at Hugging Face, so generation can be slow.

You load the model
You add a short reference clip
You type text in a supported language
The model speaks it in the cloned voice

Good to know:

• Languages: eleven Indian languages, among them Hindi, Tamil, Telugu, Bengali and Marathi.
• License: MIT
• Cost to run: none named. The online Space runs on Hugging Face hardware.
• Data it reaches: the reference audio and text you give it, and the model files it loads.
• Not for: cloning a voice without explicit permission, which the README prohibits. It is meant for research and education, and misuse is the user's responsibility.

On Hugging Face, 13 November 2026

• Likes: 1
• Built with: gradio
• License: mit

Links

• Space: https://huggingface.co/spaces/YashMachineLearning/indicF5
• Files: https://huggingface.co/spaces/YashMachineLearning/indicF5/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.