How should the dataset of PLC algorithm be constructed?


<features file>：Is it the original PCM file feature or the audio file feature after packet loss?
<packet loss file>：Is the packet loss flag corresponding to the input audio?
How should these two input files be generated?