whisper-tiny-hinglish (ONNX, quantized)

ONNX conversion of abhishekgautamm/whisper-tiny-hinglish (MIT license), a Whisper-tiny fine-tune for Hinglish (code-mixed Hindi-English) speech recognition, packaged for transformers.js v3 in the browser.

All credit for the model goes to the original author, Abhishek Gautam, and the MUCS (MUltilingual and Code-Switching) ASR challenge dataset it was trained on. This repo only converts the weights to ONNX (q8 dynamic quantization) and adds no changes of its own.

Files

onnx/encoder_model_quantized.onnx (~10 MB)
onnx/decoder_model_merged_quantized.onnx (~50 MB)

Usage (transformers.js)

import { pipeline } from '@huggingface/transformers';

const transcriber = await pipeline(
  'automatic-speech-recognition',
  'MaybeSomeone19/whisper-tiny-hinglish-onnx',
  { dtype: 'q8' },
);
const out = await transcriber(audioFloat32Array_16kHz);
console.log(out.text);

Converted with optimum-cli export onnx --task automatic-speech-recognition-with-past and quantized with onnxruntime dynamic quantization (uint8, subgraphs enabled).

Downloads last month: 44