Closed aashiqmuhamed closed 5 years ago
'data/RawNet_weights.h5' are used to extract speaker embeddings :). But, since I have already uploaded extracted speaker embeddings from entire VoxCeleb1 dataset at 'data/speaker_embeddings_RawNet_4.8eer', you can directly use those embeddings!
'data/RawNet_weights.h5' was uploaded upon a request from a researcher who wanted to extract speaker embeddings not from the VoxCeleb1 dataset but another dataset.
I'm planning on finetuning on VoxCeleb2 so I can't use your embeddings.
When I try using RawNet_weights.h5 with the CNN+GRU Rawnet model, it throws ValueError: You are trying to load a weight file containing 33 layers into a model with 38 layers.
Just making sure the weights do correspond to the net defined in model_RawNet_pre_train.py
Do you mean model_RawNet.py? 'model_RawNet_pre_train.py' refers to a CNN model which doesn't include GRU.
Can you try loading the weights one more time using 'model_RawNet.py' instead of 'model_RawNet_pre_train.py'?
Sure my bad. Thank you!
I'm wondering what network RawNet_weights.h5 provides weights for. Would you be able to provide the best model weights for both nets?