search

chenkui164 / FastASR

这是一个用C++实现ASR推理的项目，它依赖很少，安装也很简单，推理速度很快，在树莓派4B等ARM平台也可以流畅的运行。 支持的模型是由Google的Transformer模型中优化而来，数据集是开源wenetspeech(10000+小时)或阿里私有数据集(60000+小时)， 所以识别效果也很好，可以媲美许多商用的ASR软件。
Apache License 2.0
481 stars 74 forks source link

cmakelists support? #2

Closed lucasjinreal closed 2 years ago

lucasjinreal commented 2 years ago

hi. make doesn't quite support other platforms well, will consider add cmakelists build support?

chenkui164 commented 2 years ago

This suggestion is very good, but the source code uses some linux platform-specific APIs, which are not compatible with windows and macos. My plan is to implement the streaming ASR function first, and then consider multi-platform. You will be notified as soon as the modification is completed.

lucasjinreal commented 2 years ago

@chenkui164 thanks, this is a very impressive work. For Multiplatform we can using some micro definition to handle it. I didn't have a raspberrypy but I'd like test it on my Mac. Will consider deploy it into Jetsonnano

chenkui164 commented 2 years ago

The cpu of jetson nano is Quad-core ARM Cortex-A57, so it can be run directly according to the tutorial of Raspberry Pi. But the program does not use GPU, so it does not reflect the advantages of jetson nano @jinfagang

chenkui164 commented 2 years ago

Hello, the cmake version of the program has been modified and is compatible with MacOS. thanks for using。 @jinfagang