speech_recognition/FunASR
1
0
Fork 0
mirror of https://github.com/modelscope/FunASR synced 2025-09-15 14:48:36 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

readme

Browse Source
This commit is contained in:
游雁 2023-03-27 19:07:16 +08:00
parent d2d17d3ff4
commit 24932e7d76
6 changed files with 14 additions and 8441 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
funasr/runtime/grpc/Readme.md
View File

@ -52,3 +52,7 @@ Usage: ./cmake/build/paraformer_server port thread_num /path/to/model_file quant
cd ../python/grpc
python grpc_main_client_mic.py --host $server_ip --port 10108
```
## Acknowledge
1. This project is maintained by [FunASR community](https://github.com/alibaba-damo-academy/FunASR).
2. We acknowledge [DeepScience](https://www.deepscience.cn) for contributing the grpc service.

1
funasr/runtime/onnxruntime/models/readme.md
View File

@ -1 +0,0 @@
Place model.onnx here!

8404
funasr/runtime/onnxruntime/models/vocab.txt
View File

File diff suppressed because it is too large Load Diff

30
funasr/runtime/onnxruntime/readme.md
View File

@ -12,7 +12,7 @@ See the bottom of this page: Building Guidance
### 运行程序
tester /path/to/models/dir /path/to/wave/file quantize(true or false)
tester /path/to/models_dir /path/to/wave_file quantize(true or false)
例如： tester /data/models /data/test.wav false
@ -75,33 +75,11 @@ onnxruntime_xxx
└───lib
```
## 线程数与性能关系
测试环境Rocky Linux 8，仅测试cpp版本结果（未测python版本），@acely
简述：
在3台配置不同的机器上分别编译并测试，在fftw和onnxruntime版本都相同的前提下，识别同一个30分钟的音频文件，分别测试不同onnx线程数量的表现。
![线程数关系](images/threadnum.png "Windows ASR")
目前可以总结出大致规律：
并非onnx线程数越多越好
2线程比1线程提升显著，线程再多则提升较小
线程数等于CPU物理核心数时效率最好
实操建议：
大部分场景用3-4线程性价比最高
低配机器用2线程合适
## 演示
![Windows演示](images/demo.png "Windows ASR")
## 注意
本程序只支持 采样率16000hz, 位深16bit的 **单声道** 音频。
## Acknowledge
1. We acknowledge [mayong](https://github.com/RapidAI/RapidASR/tree/main/cpp_onnx) for contributing the onnxruntime(cpp api).
2. We borrowed a lot of code from [FastASR](https://github.com/chenkui164/FastASR) for audio frontend and text-postprocess.
1. This project is maintained by [FunASR community](https://github.com/alibaba-damo-academy/FunASR).
2. We acknowledge [mayong](https://github.com/RapidAI/RapidASR/tree/main/cpp_onnx) for contributing the onnxruntime(cpp api).
3. We borrowed a lot of code from [FastASR](https://github.com/chenkui164/FastASR) for audio frontend and text-postprocess.

4
funasr/runtime/python/libtorch/README.md
View File

@ -53,6 +53,10 @@
print(result)
```
## Performance benchmark
Please ref to [benchmark](https://github.com/alibaba-damo-academy/FunASR/blob/main/funasr/runtime/python/benchmark_libtorch.md)
## Speed
Environment：Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz

12
funasr/runtime/python/onnxruntime/README.md
View File

@ -54,17 +54,9 @@ python setup.py install
print(result)
```
## Speed
Environment：Intel(R) Xeon(R) Platinum 8163 CPU @ 2.50GHz
Test [wav, 5.53s, 100 times avg.](https://isv-data.oss-cn-hangzhou.aliyuncs.com/ics/MaaS/ASR/test_audio/asr_example_zh.wav)
| Backend | RTF |
|:-------:|:-----------------:|
| Pytorch | 0.110 |
| Onnx | 0.038 |
## Performance benchmark
Please ref to [benchmark](https://github.com/alibaba-damo-academy/FunASR/blob/main/funasr/runtime/python/benchmark_onnx.md)
## Acknowledge
1. This project is maintained by [FunASR community](https://github.com/alibaba-damo-academy/FunASR).