Made for Indian languages
Native script support includes Devanagari and Gurmukhi, plus Indian English. Romanized Hindi/Punjabi may require manual script conversion.
Browser-based speech generation for Hindi, Punjabi and Indian English—using an open-source experimental engine. Designed for teachers, students and content creators.
Native script support includes Devanagari and Gurmukhi, plus Indian English. Romanized Hindi/Punjabi may require manual script conversion.
The Svāra browser engine generates audio using WebGPU, Transformers.js and an SNAC decoder. The upstream implementation exports WAV.
Inference happens locally after the model downloads. Your text is not sent to a PEARL inference API. Model downloads still contact the configured model hosts.
No. The engine performs inference in compatible browsers. Hosting and model delivery may still create bandwidth costs.
Yes, the PHP host serves the webpage and compiled JavaScript. WebGPU inference runs on each visitor's device, not on Hostinger.
The engine lists Hindi, Punjabi and Indian English. Accurate Romanized language processing is not independently verified; use Hindi Devanagari or Punjabi Gurmukhi for the first tests.
The upstream browser demonstration exports WAV. Emotion tags are described upstream but results are voice- and text-dependent. MP3 requires a separate encoder integration, and is not claimed as implemented in this package.
The upstream repository identifies Apache 2.0. Before commercial launch, review the exact model-weight, tokenizer, codec and dependency licenses and preserve required notices. No blanket intellectual-property guarantee is given.