readme

游雁

2023-03-27 24932e7d7622efc1d74dd66eef6b51a972658d47

 funasr/runtime/python/onnxruntime/README.md

@@ -1,64 +1,63 @@
## Using paraformer with ONNXRuntime
## Using funasr with ONNXRuntime

<p align="left">
    <a href=""><img src="https://img.shields.io/badge/Python->=3.7,<=3.10-aff.svg"></a>
    <a href=""><img src="https://img.shields.io/badge/OS-Linux%2C%20Win%2C%20Mac-pink.svg"></a>
</p>

### Introduction
- Model comes from [speech_paraformer](https://www.modelscope.cn/models/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/summary).


### Steps:
1. Download the whole directory (`funasr/runtime/python/onnxruntime`) to the local.
2. Install the related packages.
   ```bash
   pip install requirements.txt
   ```
3. Download the model.
    - [Download Link](https://swap.oss-cn-hangzhou.aliyuncs.com/zhifu.gzf/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/model.onnx?OSSAccessKeyId=LTAI4FxMqzhBUx5XD4mKs296&Expires=2036094510&Signature=agmtMkxLEviGg3Rt3gOO4PvfrJY%3D)
    - Put the model into the `resources/models`.
        ```text
        .
        ├── demo.py
        ├── rapid_paraformer
        │   ├── __init__.py
        │   ├── kaldifeat
        │   ├── __pycache__
        │   ├── rapid_paraformer.py
        │   └── utils.py
        ├── README.md
        ├── requirements.txt
        ├── resources
        │   ├── config.yaml
        │   └── models
        │       ├── am.mvn
        │       ├── model.onnx  # Put it here.
        │       └── token_list.pkl
        ├── test_onnx.py
        ├── tests
        │   ├── __pycache__
        │   └── test_infer.py
        └── test_wavs
            ├── 0478_00017.wav
            └── asr_example_zh.wav
        ```
4. Run the demo.
1. Export the model.
   - Command: (`Tips`: torch >= 1.11.0 is required.)

       More details ref to ([export docs](https://github.com/alibaba-damo-academy/FunASR/tree/main/funasr/export))

       - `e.g.`, Export model from modelscope
         ```shell
         python -m funasr.export.export_model --model-name damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx --quantize False
         ```
       - `e.g.`, Export model from local path, the model'name must be `model.pb`.
         ```shell
         python -m funasr.export.export_model --model-name ./damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch --export-dir ./export --type onnx --quantize False
         ```


2. Install the `funasr_onnx`

install from pip
```shell
pip install --upgrade funasr_onnx -i https://pypi.Python.org/simple
```

or install from source code

```shell
git clone https://github.com/alibaba/FunASR.git && cd FunASR
cd funasr/runtime/python/funasr_onnx
python setup.py build
python setup.py install
```

3. Run the demo.
   - Model_dir: the model path, which contains `model.onnx`, `config.yaml`, `am.mvn`.
   - Input: wav formt file, support formats: `str, np.ndarray, List[str]`
   - Output: `List[str]`: recognition result.
   - Example:
        ```python
        from rapid_paraformer import RapidParaformer
        from funasr_onnx import Paraformer

        model_dir = "/nfs/zhifu.gzf/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch"
        model = Paraformer(model_dir, batch_size=1)

        config_path = 'resources/config.yaml'
        paraformer = RapidParaformer(config_path)
        wav_path = ['/nfs/zhifu.gzf/export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/example/asr_example.wav']

        wav_path = ['test_wavs/0478_00017.wav']

        result = paraformer(wav_path)
        result = model(wav_path)
        print(result)
        ```

## Performance benchmark

Please ref to [benchmark](https://github.com/alibaba-damo-academy/FunASR/blob/main/funasr/runtime/python/benchmark_onnx.md)

## Acknowledge
1. We acknowledge [SWHL](https://github.com/RapidAI/FunASR) for contributing the onnxruntime(pthon api).
1. This project is maintained by [FunASR community](https://github.com/alibaba-damo-academy/FunASR).
2. We acknowledge [SWHL](https://github.com/RapidAI/RapidASR) for contributing the onnxruntime (for paraformer model).