whisper-small-pld-fil-ONNX

ONNX / Transformers.js export of sapinsapin/whisper-small-pld-fil.

Runs fully in the browser (WebGPU or WASM) via Transformers.js. fp32 and q8 (*_quantized.onnx) weights included; q8 is ~287 MB total.

import { pipeline } from '@huggingface/transformers';
const asr = await pipeline('automatic-speech-recognition', 'internetoftim/whisper-small-pld-fil-ONNX', { dtype: 'q8', device: 'wasm' });
const out = await asr(audioFloat32Array);

Feature-extractor config taken from openai/whisper-small; tokenizer from the source repo. Exported with optimum-cli export onnx, decoders merged and dynamically quantized (QUInt8, weights only).