---
license: apache-2.0
language:
- zh
metrics:
- accuracy
- cer
pipeline_tag: automatic-speech-recognition
tags:
- Paraformer
- FunASR
- ASR
---
## Introduce
[Paraformer](https://arxiv.org/abs/2206.08317) is a non-autoregressive end-to-end speech recognition model. Compared to the currently mainstream autoregressive models, non-autoregressive models can output the target text for the entire sentence in parallel, making them particularly suitable for parallel inference using GPUs. Paraformer is currently the first known non-autoregressive model that can achieve the same performance as autoregressive end-to-end models on industrial-scale data. When combined with GPU inference, it can improve inference efficiency by 10 times, thereby reducing machine costs for speech recognition cloud services by nearly 10 times.

This repo shows how to use Paraformer with `funasr_onnx` runtime, the model comes from [FunASR](https://github.com/alibaba-damo-academy/FunASR), which trained from 60000 hours Mandarin data. The performance of Paraformer obtained the first place in [SpeechIO Leadboard](https://github.com/SpeechColab/Leaderboard).

We have released a large number of industrial-level models, including speech recognition, voice activaty detection, punctuation restoration, speaker verification, speaker diarizatio and timestamp prediction(force alignment). If you are interest, please ref to [FunASR](https://github.com/alibaba-damo-academy/FunASR).


## Install `funasr_onnx`

```shell
pip install -U funasr_onnx
# For the users in China, you could install with the command:
# pip install -U funasr_onnx -i https://mirror.sjtu.edu.cn/pypi/web/simple
```

## Download the model

```shell
git clone https://huggingface.co/funasr/paraformer-large
```

## Inference with runtime

### Speech Recognition
#### Paraformer
 ```python
 from funasr_onnx import Paraformer

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

 wav_path = ['./export/damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch/example/asr_example.wav']

 result = model(wav_path)
 print(result)
 ```
- `model_dir`: the model path, which contains `model.onnx`, `config.yaml`, `am.mvn`
- `batch_size`: `1` (Default), the batch size duration inference
- `device_id`: `-1` (Default), infer on CPU. If you want to infer with GPU, set it to gpu_id (Please make sure that you have install the onnxruntime-gpu)
- `quantize`: `False` (Default), load the model of `model.onnx` in `model_dir`. If set `True`, load the model of `model_quant.onnx` in `model_dir`
- `intra_op_num_threads`: `4` (Default), sets the number of threads used for intraop parallelism on CPU

Input: wav formt file, support formats: `str, np.ndarray, List[str]`

Output: `List[str]`: recognition result




## Performance benchmark

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