Introduction

Files in this directory are from https://modelscope.cn/models/dengcunqin/speechparaformer-largeasr_nat-zh-cantonese-en-16k-vocab8501-online/summary

First, we use the following Python code (using FunASR v0.8.8)

from funasr_onnx.paraformer_online_bin import Paraformer

model_dir = "dengcunqin/speech_paraformer-large_asr_nat-zh-cantonese-en-16k-vocab8501-online"
chunk_size = [5, 10, 5]

model = Paraformer(model_dir, batch_size=1, quantize=True, chunk_size=chunk_size, intra_op_num_threads=4) #, revision='v1.0.0') # only support batch_size = 1

to get the following files:

• model.onnx
• model_quant.onnx
• decoder.onnx
• decoder_quant.onnx

Then we rename the following three files:

• model.onnx -> encoder.onnx
• model_quant.onnx -> encoder.int8.onnx
• decoder_quant.onnx -> decoder.int8.onnx

Then we run

./generate-tokens.py

to get tokens.txt.

Finally, we run

./add-model-metadata.py

to add metadata to encoder.onnx and encoder.int8.onnx.

Now you can use the generated files in sherpa-onnx.