Open roger-tseng opened 1 year ago
@tiagoCuervo can you share the setup?
Hi there! there was a config option missing to train SCPC. I just pushed a commit fixing that, and the instructions to run an SCPC training session are now in the README.
Hello,
I followed the instructions for running SCPC. Below is the command used
python cpc/train.py --pathDB /path_datasets/LibriSpeech/train-clean-100 --file_extension '.flac' --pathCheckpoint ./scpc_test --normMode layerNorm --dropout --n_process_loader 1 --batchSizeGPU 32 --nPredicts 1 --limitNegsInBatch 8 --nEpoch 50 --nGPU 1 --schedulerRamp 10 --rnnMode none --arMode no_ar --negativeSamplingExt 1 --nPredictsSegment 1 --samplingType samesequence --linearOutput --normalizeCPCScore --multiLevel --segmentationMode cosineDissimilarity --rnnModeSegment none
However, I get the following error:
File "/export/hCPC/cpc/criterion/criterion.py", line 250, in forward
batchSize, seqSize, dimAR = cFeature.size()
AttributeError: 'list' object has no attribute 'size'
Could you please take a look at the error? Thank you.
Hello! I would like to compare your method's phoneme segmentation results with SCPC, but wasn't sure about the specific arguments. Could you share your training script for SCPC?