Atlas 800I A2部署funASR记录_昇腾主版块_昇腾论坛

# Atlas 800I A2部署funASR记录

### 精华推荐

发表于 2025-08-08 09:27:2416查看

1.  **拉取镜像**
    
    `docker pull swr.cn-southwest-2.myhuaweicloud.com/atelier/pytorch_2_1_ascend:pytorch_2.1.0-cann_8.0.rc3-py_3.9-hce_2.0.2406-aarch64-snt9b-20240910112800-2a95df3`
    
2.  **启动镜像**
    
    `docker run -itd -u=0 --net=host \ --device=/dev/davinci0 \ --device=/dev/davinci1 \ --device=/dev/davinci_manager \ --device=/dev/devmm_svm \ --device=/dev/hisi_hdc \ --shm-size=32g \ -v /usr/local/dcmi:/usr/local/dcmi \ -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \ -v /var/log/npu/:/usr/slog \ -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \ -v $(pwd):/work \ --name funasr \ 243888a362d3 \ /bin/bash`
    
3.  **安装第三方依赖**
    
     `pip install funasr==1.1.12 torchaudio==2.1.0 tqdm==4.66.6`
    
4.  **下载模型文件**
    
    `modelscope download --model iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch --local_dir ./speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch`
    
5.  **下载测试数据集** 中文语音识别Aishell-1学术数据集测试集 · 数据集
6.  **创建推理脚本**
    
    `from funasr import AutoModel import torch_npu from torch_npu.contrib import transfer_to_npu  path_asr  = 'speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch'  model = AutoModel(     model               = path_asr,     model_revision      = "v2.0.4",     device              = "cuda:0", )  #warmup warmup_file="/work/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch/example/asr_example.wav" text = model.generate(input=warmup_file)[0]["text"] print(text)  input_files = "aishell.scp" result_text = model.generate(input=input_files)  with open(input_files, 'r') as file:     lines = file.readlines() first_columns = [line.split()[0] for line in lines]  result_file = "infer_result.txt" with open(result_file, 'w') as file:         for i in range(len(result_text)):             file.write(str(first_columns[i]) + ' ' + str(result_text[i]["text"]) + '\n')             print(first_columns[i]+' '+result_text[i]["text"])`
    
7.  **编写aishell.scp**
    
    `BAC009S0764W0121 speech_asr_aishell1_testsets/wav/test/S0764/BAC009S0764W0121.wav BAC009S0764W0122 speech_asr_aishell1_testsets/wav/test/S0764/BAC009S0764W0122.wav BAC009S0764W0123 speech_asr_aishell1_testsets/wav/test/S0764/BAC009S0764W0123.wav BAC009S0764W0124 speech_asr_aishell1_testsets/wav/test/S0764/BAC009S0764W0124.wav ...`
    
8.  **执行测试脚本输出结果**![image-20250806161420259.png](https://yg9538.kmgy.top/img/2025/08/26/eeba472f5f3a4762aafd12525dce5d954600.png)