wav2vec2-large-xlsr-53-french-ONNX
ONNX export of jonatasgrosman/wav2vec2-large-xlsr-53-french for char-CTC
forced alignment / ASR (wav2vec2 family).
These weights were prepared for the mobile SDK react-native-sherpa-onnx (offline speech tooling for React Native / Android / iOS), including forced-alignment model packs used by the app.
Try it live in VoiceLab
Download and run these on-device models in your pocket — anytime, anywhere — in the mobile app VoiceLab:
- Android: VoiceLab on Google Play
- iOS: VoiceLab on the App Store
- Product page: https://xdcobra.github.io/voice-lab-offline-tools/
Files
| Path | Notes |
|---|---|
onnx/model.onnx |
fp32 |
onnx/model_fp16.onnx |
fp16 weights (when present) |
onnx/model_int8.onnx |
dynamic QUInt8 weight quantization |
vocab.json |
CTC character vocabulary |
config.json |
Transformers model config |
LICENSE |
apache-2.0 |
Source
- Base checkpoint:
jonatasgrosman/wav2vec2-large-xlsr-53-french(apache-2.0) - Export:
optimum-cli export onnx+onnxruntimedynamic quantization (export_hf_onnx_pack.pyin the SDK repo) - Intended use: offline forced alignment (CTC) in commercial apps when the base model license allows it
Sample rate
16 kHz mono PCM.