Paraformer-large模型介绍

使用方法

下载模型

git clone https://github.com/manyeyes/AliParaformerAsr.git
cd /AliParaformerAsr/AliParaformerAsr.Examples
git clone https://huggingface.co/manyeyes/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx.git

编辑 Program.cs 中的 modelName 值：

string modelName="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx";
//[fp32模型/int8模型]，二选一
//fp32模型
//string modelFilePath = applicationBase + "./"+ modelName + "/model.onnx";
//int8模型
string modelFilePath = applicationBase + "./"+ modelName + "/model_quant.onnx";

将文件夹 speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx 中的文件属性-》复制到输出目录-》如果较新则复制。 或者 编辑 AliParaformerAsr.Examples.csproj ，添加

<ItemGroup>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\am.mvn">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\asr.yaml">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\example\0.wav">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\example\1.wav">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\example\2.wav">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\example\3.wav">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\example\4.wav">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\model_quant.onnx">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\model.onnx">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
    <None Update="speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx\tokens.txt">
      <CopyToOutputDirectory>PreserveNewest</CopyToOutputDirectory>
    </None>
  </ItemGroup>

启动 AliParaformerAsr.Examples

通过modelscope了解更多

原模型

• pytorch https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary
• onnx https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-onnx/summary

Highlights

• 热词版本：Paraformer-large热词版模型支持热词定制功能，基于提供的热词列表进行激励增强，提升热词的召回率和准确率。
• 长音频版本：Paraformer-large长音频模型，集成VAD、ASR、标点与时间戳功能，可直接对时长为数小时音频进行识别，并输出带标点文字与时间戳。
• 基于FunASR进行模型的微调和推理： FunASR框架支持魔搭社区开源的工业级的语音识别模型的training & finetuning，使得研究人员和开发者可以更加便捷的进行语音识别模型的研究和生产，目前已在Github开源：https://github.com/alibaba-damo-academy/FunASR 。