phase1.json

[{"training_name": "[2019-05-01 15:56:12]: [190501_1556_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(1.000)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(1.000)", "proc_func_label": "s_to_exp(1.000)"}}, "model": {"name": "190501_1556_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(1.000)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 15:58:56]: [190501_1558_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(1.000)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(1.000)", "proc_func_label": "s_to_exp(1.000)"}}, "model": {"name": "190501_1558_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(1.000)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 16:16:43]: [190501_1616_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(2.000)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(2.000)", "proc_func_label": "s_to_exp(2.000)"}}, "model": {"name": "190501_1616_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(2.000)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 16:31:06]: [190501_1631_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(0.167)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(0.167)", "proc_func_label": "s_to_exp(0.167)"}}, "model": {"name": "190501_1631_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(0.167)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 16:52:39]: [190501_1652_conv3_small_norm_8d6ce6 conv3_small_norm.jsont] -> [s_to_reim_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_reim", "proc_func_label": "s_to_reim"}}, "model": {"name": "190501_1652_conv3_small_norm_8d6ce6", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_reim_norm.h5", "input_shape": [256, 32, 2], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 512, "timesteps": 32, "channels": 2, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 2)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      4864      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 2)        4610      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 2)        0         \n=================================================================\nTotal params: 328,226\nTrainable params: 326,882\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 2, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 17:28:46]: [190501_1728_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_db_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_db", "proc_func_label": "s_to_db"}}, "model": {"name": "190501_1728_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_db_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 17:44:45]: [190501_1744_conv3_small_582d81 conv3_small.jsont] -> [s_to_exp(1.000)_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(1.000)", "proc_func_label": "s_to_exp(1.000)"}}, "model": {"name": "190501_1744_conv3_small_582d81", "source": "models/conv3_small.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(1.000)_no.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "zeros"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 18:05:15]: [190501_1805_conv3_small_582d81 conv3_small.jsont] -> [s_to_exp(2.000)_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(2.000)", "proc_func_label": "s_to_exp(2.000)"}}, "model": {"name": "190501_1805_conv3_small_582d81", "source": "models/conv3_small.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(2.000)_no.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "zeros"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 18:19:05]: [190501_1819_conv3_small_582d81 conv3_small.jsont] -> [s_to_exp(0.167)_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(0.167)", "proc_func_label": "s_to_exp(0.167)"}}, "model": {"name": "190501_1819_conv3_small_582d81", "source": "models/conv3_small.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(0.167)_no.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "zeros"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 18:51:46]: [190501_1851_conv3_small_norm_8d6ce6 conv3_small_norm.jsont] -> [s_to_reim_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_reim", "proc_func_label": "s_to_reim"}}, "model": {"name": "190501_1851_conv3_small_norm_8d6ce6", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_reim_no.h5", "input_shape": [256, 32, 2], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 512, "timesteps": 32, "channels": 2, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 2)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      4864      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 2)        4610      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 2)        0         \n=================================================================\nTotal params: 328,226\nTrainable params: 326,882\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 2, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-01 19:27:49]: [190501_1927_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_db_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_db", "proc_func_label": "s_to_db"}}, "model": {"name": "190501_1927_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_db_no.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-02 01:00:04]: [190502_0100_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(0.667)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(0.667)", "proc_func_label": "s_to_exp(0.667)"}}, "model": {"name": "190502_0100_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(0.667)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-02 01:01:05]: [190502_0101_conv3_small_norm_d2f153 conv3_small_norm.jsont] -> [s_to_exp(0.667)_norm.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(0.667)", "proc_func_label": "s_to_exp(0.667)"}}, "model": {"name": "190502_0101_conv3_small_norm_d2f153", "source": "models/conv3_small_norm.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(0.667)_norm.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "ones"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "ones"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}, {"training_name": "[2019-05-02 01:31:10]: [190502_0131_conv3_small_582d81 conv3_small.jsont] -> [s_to_exp(0.667)_no.h5]", "data": {"clean": {"dataset_path": "/data/riccardo_datasets/npr_news/ds0/train", "filepath_list_train": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-15-2018_11PM_ET.wav"], "filepath_list_valid": ["/data/riccardo_datasets/npr_news/ds0/train/NPR_News__03-16-2018_11PM_ET.wav"]}, "noise": {"rir_path": null, "noise_snrs": [15], "noise_funcs": ["pink_noise"]}, "processing": {"sr": 16000, "n_fft": 512, "hop_length": 128, "win_length": 512, "frag_hop_length": 30, "frag_win_length": 32, "proc_func": "s_to_exp(0.667)", "proc_func_label": "s_to_exp(0.667)"}}, "model": {"name": "190502_0131_conv3_small_582d81", "source": "models/conv3_small.jsont", "destination": "/data/riccardo_models/denoising/phase1/s_to_exp(0.667)_no.h5", "input_shape": [256, 32, 1], "template_args": {"n_filters": 256, "n_conv": 256, "n_recurrent": 512, "ker_size": 3, "n_dense": 256, "timesteps": 32, "channels": 1, "dropout_rate": 0.0, "activ_func": "relu", "n_stacks": 8, "dilations": [1, 2], "use_skip_connections": "true", "return_sequences": "true", "bias_initializer": "zeros", "strides": [2, 2]}, "time_slice": [null, null], "summary": "_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\ninput_1 (InputLayer)         (None, 256, 32, 1)        0         \n_________________________________________________________________\nconv2d_1 (Conv2D)            (None, 128, 16, 256)      2560      \n_________________________________________________________________\nbatch_normalization_1 (Batch (None, 128, 16, 256)      1024      \n_________________________________________________________________\nactivation_1 (Activation)    (None, 128, 16, 256)      0         \n_________________________________________________________________\nconv2d_2 (Conv2D)            (None, 64, 8, 64)         147520    \n_________________________________________________________________\nbatch_normalization_2 (Batch (None, 64, 8, 64)         256       \n_________________________________________________________________\nactivation_2 (Activation)    (None, 64, 8, 64)         0         \n_________________________________________________________________\nconv2d_3 (Conv2D)            (None, 32, 4, 16)         9232      \n_________________________________________________________________\nbatch_normalization_3 (Batch (None, 32, 4, 16)         64        \n_________________________________________________________________\nactivation_3 (Activation)    (None, 32, 4, 16)         0         \n_________________________________________________________________\nconv2d_transpose_1 (Conv2DTr (None, 64, 8, 16)         2320      \n_________________________________________________________________\nbatch_normalization_4 (Batch (None, 64, 8, 16)         64        \n_________________________________________________________________\nactivation_4 (Activation)    (None, 64, 8, 16)         0         \n_________________________________________________________________\nconv2d_transpose_2 (Conv2DTr (None, 128, 16, 64)       9280      \n_________________________________________________________________\nbatch_normalization_5 (Batch (None, 128, 16, 64)       256       \n_________________________________________________________________\nactivation_5 (Activation)    (None, 128, 16, 64)       0         \n_________________________________________________________________\nconv2d_transpose_3 (Conv2DTr (None, 256, 32, 256)      147712    \n_________________________________________________________________\nbatch_normalization_6 (Batch (None, 256, 32, 256)      1024      \n_________________________________________________________________\nactivation_6 (Activation)    (None, 256, 32, 256)      0         \n_________________________________________________________________\nconv2d_transpose_4 (Conv2DTr (None, 256, 32, 1)        2305      \n_________________________________________________________________\nactivation_7 (Activation)    (None, 256, 32, 1)        0         \n=================================================================\nTotal params: 323,617\nTrainable params: 322,273\nNon-trainable params: 1,344\n_________________________________________________________________\n", "architecture": [{"layer_type": "conv2d", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2d", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 16, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 64, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 256, "kernel_size": 3, "padding": "same", "strides": [2, 2], "bias_initializer": "zeros"}}, {"layer_type": "batchnorm", "layer_args": {}}, {"layer_type": "activation", "layer_args": {"activation": "relu"}}, {"layer_type": "conv2dt", "layer_args": {"filters": 1, "kernel_size": 3, "padding": "same", "strides": 1, "bias_initializer": "zeros"}}, {"layer_type": "activation", "layer_args": {"activation": "linear"}}], "description": {"desc": " AE with 3 convolutional layers, used for sanity check", "source": "models/conv3.jsont"}}, "training": {"epochs": 500, "initial_epoch": 0, "max_epochs": 500, "batch_size": 128, "train_steps_per_epoch": 9, "valid_steps_per_epoch": 9, "cuda_device": "0", "learning_rate": {"initial_lr": 0.005, "lr_drop_rate": 0.5, "lr_drop_epochs": 150}}}]