Modelpublic

dmouayad/Audar-ASR-V1-Flash-sherpa-onnx

sourceHugging Faceotherupdated 26d agoView on Hugging Face

0likes52downloads

Model Card

Audar-ASR-V1-Flash (sherpa-onnx)

ONNX export of Audar-ASR-V1-Flash for use with sherpa-onnx and the reference infer_qwen3_asr.py script.

Files included

File	Description	Size	Precision
`conv_frontend.onnx`	Conv frontend	43 MB	FP32
`encoder.onnx` + `.data`	Encoder	~675 MB	FP32
`decoder.int8.onnx`	Decoder (quantized)	~720 MB	INT8
`encoder.int8.onnx`	Alternative quantized encoder	175 MB	INT8
`decoder.onnx` + `.data`	Original FP32 decoder (optional)	~2.8 GB	FP32
`config.json`, `tokenizer.json`, `tokenizer_config.json`, `vocab.json`	Tokenizer & model config	—	—

Recommended usage (best quality/size balance)

bash

python3 infer_qwen3_asr.py \
    --model . \
    --conv_frontend conv_frontend.onnx \
    --encoder encoder.onnx \
    --decoder decoder.int8.onnx \
    --wav your_audio.wav \
    --device cpu

Best quality: FP32 encoder + INT8 decoder

Source

Original model: Audar-ASR-V1-Flash

Export script: https://github.com/Wasser1462/Qwen3-ASR-onnx