Meta Omnilingual ASR v2 (ONNX Quantized)

This repository contains the ONNX INT8/FP32 quantized versions of Meta AI's Omnilingual ASR v2 models, supporting over 1600 languages.

Converted and quantized by Edison dos Santos for use with sherpa-onnx.

Files

The following archives include the model, tokens, license, and test audio:

Archive File	Size	SHA256 Hash
`sherpa-onnx-omnilingual-asr-1600-languages-1B-ctc-v2-int8-2026-02-05.tar.bz2`	~688 MB	`2872aacc3d91b2be3d00623576025b64998d7ab0e3a463e40b8839b5e673a631`
`sherpa-onnx-omnilingual-asr-1600-languages-1B-ctc-v2-int4-2026-02-06.tar.bz2`	~525 MB	`6bff0f0b399ce5d1e8a79e389d5ad6ee4e4250f745d297f0d724af8ca0d2ac3e`
`sherpa-onnx-omnilingual-asr-1600-languages-300M-ctc-v2-2026-02-05.tar.bz2`	~441 MB	`7d5820c86690fc90882c67705fe2dd2f63f1318e46f7421c055655894b0a213f`
`sherpa-onnx-omnilingual-asr-1600-languages-300M-ctc-v2-int8-2026-02-05.tar.bz2`	~235 MB	`b72bef9be75862684098e722d79fefecf9f10fefd3a0b2950738977b4c6b4147`
`sherpa-onnx-omnilingual-asr-1600-languages-300M-ctc-v2-int4-2026-02-06.tar.bz2`	~187 MB	`f3a02e7df68e3648be9c591c0b111d76ac814e16b61a7be8c77ffd0d10028918`

./sherpa-onnx-offline \
  --omnilingual-asr-model=./model.int8.onnx \
  --tokens=./tokens.txt \
  ./test_wavs/en.wav

Downloads last month: -; Downloads are not tracked for this model. How to track