1*a58d3d2aSXin Li#Packet loss simulator 2*a58d3d2aSXin Li 3*a58d3d2aSXin LiThis code is an attempt at simulating better packet loss scenarios. The most common way of simulating 4*a58d3d2aSXin Lipacket loss is to use a random sequence where each packet loss event is uncorrelated with previous events. 5*a58d3d2aSXin LiThat is a simplistic model since we know that losses often occur in bursts. This model uses real data 6*a58d3d2aSXin Lito build a generative model for packet loss. 7*a58d3d2aSXin Li 8*a58d3d2aSXin LiWe use the training data provided for the Audio Deep Packet Loss Concealment Challenge, which is available at: 9*a58d3d2aSXin Li 10*a58d3d2aSXin Lihttp://plcchallenge2022pub.blob.core.windows.net/plcchallengearchive/test_train.tar.gz 11*a58d3d2aSXin Li 12*a58d3d2aSXin LiTo create the training data, run: 13*a58d3d2aSXin Li 14*a58d3d2aSXin Li`./process_data.sh /<path>/test_train/train/lossy_signals/` 15*a58d3d2aSXin Li 16*a58d3d2aSXin LiThat will create an ascii loss\_sorted.txt file with all loss data sorted in increasing packet loss 17*a58d3d2aSXin Lipercentage. Then just run: 18*a58d3d2aSXin Li 19*a58d3d2aSXin Li`python ./train_lossgen.py` 20*a58d3d2aSXin Li 21*a58d3d2aSXin Lito train a model 22*a58d3d2aSXin Li 23*a58d3d2aSXin LiTo generate a sequence, run 24*a58d3d2aSXin Li 25*a58d3d2aSXin Li`python3 ./test_lossgen.py <checkpoint> <percentage> output.txt --length 10000` 26*a58d3d2aSXin Li 27*a58d3d2aSXin Liwhere <checkpoint> is the .pth model file and <percentage> is the amount of loss (e.g. 0.2 for 20% loss). 28