Ninada

Ninada

Reads text aloud in 12 languages, including Sanskrit, Indian English, Hindi and Tamil, with 24 kHz voices that run in a browser or on a phone. Two ONNX tiers, one for desktops and one for mobile.

Made by gnumanth on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Ninada is a text-to-speech suite with 24 kHz voices for 12 languages, each with a male and a female voice. That makes 24 voices and 48 ONNX models. A studio tier suits desktops, and a smaller mobile tier suits phones and low-memory CPUs.

Who it suits: It suits developers who add speech to web pages or apps, especially for Indian languages and Sanskrit. It also suits researchers who compare lightweight voice models.

Title: Who runs it
Industry | Who uses it | How they use it | Example
Software | Web and app developers | Add spoken output in a browser or on a phone | Read a page aloud in Hindi
Education | Language learners and teachers | Hear Sanskrit or other Indian languages spoken | Listen to a Sanskrit verse
Research | Speech researchers | Compare studio and mobile vocoder tiers | Test the two tiers on the same text

If you need voices for languages outside the 12, you can skip it.

Where to use it:

• A browser, through the project's web studio, with WebGPU on desktops and WebAssembly on phones
• Desktop machines, which load the studio files
• Mobile browsers and low-memory CPUs, which load the mobile files
• Streaming kits in JavaScript and Python, listed in the author's GitHub repository

Title: Jobs it does
Use case | Who | What they get back
Speak text on a desktop | Developers | 24 kHz audio from the studio voices
Speak text on a phone | Mobile developers | 24 kHz audio streamed from the lighter mobile voices
Multi-language voice apps | App teams | A male and a female voice in each of 12 languages

How to set it up:

• Python, to download the files by hand
• An ONNX-capable runtime, or the author's JavaScript or Python streaming kits
• Studio files from the onnx folder, or mobile files from the onnx_mobile folder

1. Install the Hugging Face Hub tools.
2. Download the repository files.
3. Load the studio or mobile ONNX file for your language and voice in your runtime.

pip install -U huggingface_hub
hf download gnumanth/ninada

How to use it: Open the project's web studio in a browser, or load a voice file in your own runtime. Desktops and WebGPU clients take the studio files, and mobile browsers take the mobile files automatically. The card gives no other commands.

hf download gnumanth/ninada

Example: You choose Hindi and the male voice, then enter a sentence in the web studio. The page speaks it aloud as 24 kHz audio. On a phone, the lighter mobile voice is loaded for you, and the card reports no streaming underruns there.

For newcomers and experts: If you are new, text-to-speech turns written words into a spoken voice. The first small thing to try is the web studio in a browser, which needs no download.

If you know the field, the studio tier is a full HiFi-GAN v1 generator and the mobile tier is a distilled iSTFTNet decoder with 4.1 times fewer vocoder MACs. The card measured about 43 ms to first audio on the studio tier and 519 ms first-chunk compute on a Google Pixel 10 Pro XL. The trade-off is that mobile voices are smaller distillations of the studio ones.

Enter text in one of 12 languages
Pick a male or female voice
The runtime loads studio or mobile ONNX
The voice renders 24 kHz audio

Good to know:

Title: Size of one voice model
Studio tier: 27.93 MiB
Mobile tier: 21.68 MiB

• Languages: the card lists 12, while Hugging Face's own record shows 8 of them.
• Voices: 24 voices in all, made of 12 male and 12 female specialists.
• Training data: studio corpora named in the card carry the CC-BY-4.0 license.
• Hosting: the web studio, the SDKs and the dataset live outside this repository.
• License: cc-by-4.0.
• Cost to run: no API key, account or paid service is named.
• Data it reaches: it downloads files from Hugging Face, and the card names no other access.

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: onnx
• Parameters: 81.9M
• Weights: about 328 MB in F32; about 41 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 292
• Likes: 4
• License: cc-by-4.0
• Languages: sa, en, hi, kn, mr, bn, ta, te
• Tags: onnx, safetensors, vits, text-to-speech, streaming-tts, ninada, sanskrit, english, indic, kannada, marathi, bengali

Links

• Model: https://huggingface.co/gnumanth/ninada
• Files: https://huggingface.co/gnumanth/ninada/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download gnumanth/ninada

Then follow the usage in its model card. Nothing in this kit runs by itself.