sherpa-onnx-wenet-ax650

WeNet U2++ Conformer CTC Chinese ASR model converted to AX650 AXMODEL for on-device inference via sherpa-onnx.

Model Details

• Original Model: csukuangfj/sherpa-onnx-zh-wenet-aishell
• AXMODEL Source: AXERA-TECH/WeNet encoder
• Architecture: WeNet U2++ Conformer CTC
• Task: Chinese speech recognition (ASR)
• Sample rate: 16000 Hz
• Feature dim: 80-dim mel filterbanks
• Subsampling factor: 4
• Vocabulary size: 4233
• Input: speech (FP32, [1, T, 80]), speech_lengths (INT32, [1])
• Output: CTC log probabilities ctc_log_probs (FP32, [1, T', 4233])
• Target Chip: AX650 (NPU3)
• Quantization: U16 (whole graph)

Files

File	Size	Description
wenet-encoder.axmodel	41 MB	AX650 encoder model
tokens.txt	-	Vocabulary file (4233 tokens)

Usage with sherpa-onnx

./sherpa-onnx-offline \
  --wenet-ctc-model=wenet-encoder.axmodel \
  --tokens=tokens.txt \
  --provider=axera \
  test.wav

Performance

Metric	Value
ONNX size (original)	138 MB
AXMODEL size	41 MB
Compression ratio	3.4x

Conversion Details

• Pulsar2 Version: 6.0
• Calibration: MinMax with 10 samples
• Quantization: U16

License

Apache 2.0