Show HN: In-Browser Sanskrit ASR Model

Category: ai-ml

Tags: sanskrit, asr, onnx, quantization, speech-recognition, in-browser

Score: 5.7/10 (Innovation: 6, Technical: 7, Documentation: 6, Utility: 4)

This repository provides ONNX and INT8 quantized exports of Su-śrotā, a scholar-grade Sanskrit ASR model, enabling fully serverless in-browser speech recognition via onnxruntime-web or local deployment without PyTorch/NeMo dependencies. It packages a preprocessor, quantized Conformer-CTC model, and vocabulary, with demonstrated real-time performance on CPU and WebAssembly.

Target audience: ML engineers, Sanskrit/NLP researchers, web developers integrating speech recognition

Repository: https://huggingface.co/gnumanth/sushrota-sanskrit-asr-onnx

View on Hacker News