Launching today
Ninada
Sub-second in-browser 24kHz TTS for 12 Indian languages
2 followers
Sub-second in-browser 24kHz TTS for 12 Indian languages
2 followers
Stream 24 kHz speech privately in-browser across Sanskrit, Indian English, and 10 Indian languages. Powered by 28 MB native per-language VITS ONNX specialists (83 ms WebGPU / 554 ms low-RAM WASM) with offline caching plus Node, Python, and CLI SDKs.
Launch Team
Atlassian LoomTalk it through in Loom. Ship with AI
Promoted
Hey Product Hunt!
Most Indic TTS pipelines either require shipping user audio to a cloud API or downloading multi-gigabyte multilingual checkpoints that bleed accents across scripts and crash mobile browsers.
I built Ninada to stream 24,000 Hz studio speech synthesis directly inside a browser tab, Node.js, or Python in under a second:
- 24 dedicated 27.9 MB W8A32+FP16 VITS ONNX specialists (Male and Female across Classical Sanskrit, Indian English, Hindi, Kannada, Marathi, Bengali, Tamil, Telugu, Malayalam, Gujarati, Punjabi, and Odia) trained natively on each script with zero cross-language accent bleed.
- Sub-second tap-to-audible streaming: 83 ms on WebGPU and 554 ms on 2 GB RAM throttled mobile WASM with 0 AudioWorklet underruns.
- 162 MB total browser memory via a zero-copy 3-way ONNX graph splitter (prior + flow + vocoder) and arena-free WASM execution.
- Offline CacheStorage + IndexedDB persistence (0 network refetches after first load) plus a 0.0 MB zero-dependency Klatt fallback synthesizer.
Try it right in the browser (no sign-up, no API keys) or via npx ninada / pip install ninada:
- Live Studio: https://hemanth.github.io/ninada/
- GitHub (MIT): https://github.com/hemanth/ninada
Would love feedback on voice prosody across your native language and any low-memory devices you test on!