initial commit

c1fb5b6 almost 5 years ago

8.27 kB

	# Generated 2021-09-17 from:
	# /home/mila/s/subakany/speechbrain_new/recipes/WSJ0Mix/separation/snrestimator_yamls/timedom_convnet_whamr_v2_stnorm_manyseparators.yaml
	# yamllint disable
	# ################################
	# Model: SepFormer for source separation
	# https://arxiv.org/abs/2010.13154
	# Dataset : WSJ0-2mix and WSJ0-3mix
	# ################################
	#
	# Basic parameters
	# Seed needs to be set at top of yaml, before objects with parameters are made
	#
	seed: 1234
	__set_seed: !apply:torch.manual_seed [1234]

	# Data params

	# e.g. '/yourpath/wsj0-mix/2speakers'
	# end with 2speakers for wsj0-2mix or 3speakers for wsj0-3mix
	data_folder: /miniscratch/subakany/LibriMixData_new/Libri2Mix/

	# the path for wsj0/si_tr_s/ folder -- only needed if dynamic mixing is used
	# e.g. /yourpath/wsj0-processed/si_tr_s/
	# you need to convert the original wsj0 to 8k
	# you can do this conversion with the script ../meta/preprocess_dynamic_mixing.py
	base_folder_dm: /miniscratch/subakany/LibriMixData_new/LibriSpeech/train-clean-360_processed/
	rir_path: /miniscratch/subakany/whamr_rirs_wav

	experiment_name: snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators
	output_folder: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234
	train_log: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/train_log.txt
	save_folder: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save
	train_data: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save/libri2mix_train-360.csv
	valid_data: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save/libri2mix_dev.csv
	test_data: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save/libri2mix_test.csv

	wsj_data_folder: /network/tmp1/subakany/wham_original
	train_wsj_data: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save/wham_tr.csv
	test_wsj_data: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save/wham_tt.csv
	base_folder_dm_whamr: /network/tmp1/subakany/wsj0-processed/si_tr_s
	use_whamr_train: true
	whamr_proportion: 0.6

	test_onwsj: false

	skip_prep: false

	ckpt_interval_minutes: 60

	# Experiment params
	auto_mix_prec: false # Set it to True for mixed precision
	test_only: false
	num_spks: 2 # set to 3 for wsj0-3mix
	progressbar: true
	save_audio: false # Save estimated sources on disk
	sample_rate: 8000

	# Training parameters
	N_epochs: 200
	batch_size: 1
	lr: 0.0001
	clip_grad_norm: 5
	loss_upper_lim: 999999 # this is the upper limit for an acceptable loss
	# if True, the training sequences are cut to a specified length
	limit_training_signal_len: false
	# this is the length of sequences if we choose to limit
	# the signal length of training sequences
	training_signal_len: 32000000

	# Set it to True to dynamically create mixtures at training time
	dynamic_mixing: true
	use_wham_noise: true
	use_reverb_augment: true

	# Parameters for data augmentation
	use_wavedrop: false
	use_speedperturb: true
	use_speedperturb_sameforeachsource: false
	use_rand_shift: false
	min_shift: -8000
	max_shift: 8000

	speedperturb: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
	perturb_prob: 1.0
	drop_freq_prob: 0.0
	drop_chunk_prob: 0.0
	sample_rate: 8000
	speeds: [95, 100, 105]

	wavedrop: !new:speechbrain.lobes.augment.TimeDomainSpecAugment
	perturb_prob: 0.0
	drop_freq_prob: 1.0
	drop_chunk_prob: 1.0
	sample_rate: 8000

	# loss thresholding -- this thresholds the training loss
	threshold_byloss: true
	threshold: -30

	# Encoder parameters
	N_encoder_out: 256
	out_channels: 256
	kernel_size: 16
	kernel_stride: 8

	# Dataloader options
	dataloader_opts:
	batch_size: 1
	num_workers: 0


	# Specifying the network
	Encoder: &id003 !new:speechbrain.lobes.models.dual_path.Encoder
	kernel_size: 16
	out_channels: 256


	SBtfintra: &id001 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
	num_layers: 8
	d_model: 256
	nhead: 8
	d_ffn: 1024
	dropout: 0
	use_positional_encoding: true
	norm_before: true

	SBtfinter: &id002 !new:speechbrain.lobes.models.dual_path.SBTransformerBlock
	num_layers: 8
	d_model: 256
	nhead: 8
	d_ffn: 1024
	dropout: 0
	use_positional_encoding: true
	norm_before: true

	MaskNet: &id005 !new:speechbrain.lobes.models.dual_path.Dual_Path_Model
	num_spks: 2
	in_channels: 256
	out_channels: 256
	num_layers: 2
	K: 250
	intra_model: *id001
	inter_model: *id002
	norm: ln
	linear_layer_after_inter_intra: false
	skip_around_intra: true

	Decoder: &id004 !new:speechbrain.lobes.models.dual_path.Decoder
	in_channels: 256
	out_channels: 1
	kernel_size: 16
	stride: 8
	bias: false

	snrmin: 0
	snrmax: 10
	out_n_neurons: 16
	use_snr_compression: true
	separation_norm_type: stnorm

	# compute_features: !new:speechbrain.lobes.features.Fbank
	# n_mels: !ref <n_mels>
	# left_frames: 0
	# right_frames: 0
	# deltas: False

	latent_dim: 128
	n_inp: 256
	classifier_enc: &id006 !new:speechbrain.nnet.containers.Sequential
	input_shape: [!!null '', 2, !!null '']
	cnn1: !new:speechbrain.nnet.CNN.Conv1d
	in_channels: 2
	kernel_size: 4
	out_channels: 128
	stride: 1
	skip_transpose: true
	padding: valid
	relu1: !new:torch.nn.ReLU
	cnn2: !new:speechbrain.nnet.CNN.Conv1d
	in_channels: 128
	kernel_size: 4
	out_channels: 128
	stride: 2
	skip_transpose: true
	padding: valid
	relu2: !new:torch.nn.ReLU
	cnn3: !new:speechbrain.nnet.CNN.Conv1d
	in_channels: 128
	kernel_size: 4
	out_channels: 128
	stride: 2
	skip_transpose: true
	padding: valid
	relu3: !new:torch.nn.ReLU
	cnn4: !new:speechbrain.nnet.CNN.Conv1d
	in_channels: 128
	kernel_size: 4
	out_channels: 128
	stride: 2
	skip_transpose: true
	padding: valid
	relu4: !new:torch.nn.ReLU
	cnn5: !new:speechbrain.nnet.CNN.Conv1d
	in_channels: 128
	kernel_size: 4
	out_channels: 128
	stride: 2
	skip_transpose: true
	padding: valid

	stat_pooling: !new:speechbrain.nnet.pooling.StatisticsPooling


	# classifier_enc: !new:speechbrain.lobes.models.ECAPA_TDNN.ECAPA_TDNN
	# input_size: !ref <n_inp>
	# channels: [1024, 1024, 1024, 1024, 3072]
	# kernel_sizes: [5, 3, 3, 3, 1]
	# dilations: [1, 2, 3, 4, 1]
	# attention_channels: 128
	# lin_neurons: 192

	#classifier_out: !new:speechbrain.lobes.models.ECAPA_TDNN.Classifier
	# input_size: 192
	# out_neurons: !ref <out_n_neurons>
	#
	# classifier_out: !new:speechbrain.nnet.linear.Linear
	# input_size: 256
	# n_neurons: 1

	classifier_out: &id007 !new:speechbrain.nnet.containers.Sequential
	# lr_scheduler: !ref <lr_scheduler>

	input_shape: [!!null '', 256]
	layer1: !new:speechbrain.nnet.linear.Linear
	input_size: 256
	n_neurons: 256
	relu: !new:torch.nn.ReLU
	layer2: !new:speechbrain.nnet.linear.Linear
	input_size: 256
	n_neurons: 1
	sigm: !new:torch.nn.Sigmoid



	classifier_loss: !new:torch.nn.CrossEntropyLoss

	optimizer: !name:torch.optim.Adam
	lr: 0.0001
	weight_decay: 0

	loss: !name:speechbrain.nnet.losses.get_si_snr_with_pitwrapper

	lr_scheduler: !new:speechbrain.nnet.schedulers.ReduceLROnPlateau
	factor: 0.5
	patience: 2
	dont_halve_until_epoch: 95

	epoch_counter: &id008 !new:speechbrain.utils.epoch_loop.EpochCounter
	limit: 200

	modules:
	encoder: *id003
	decoder: *id004
	masknet: *id005
	classifier_enc: *id006
	classifier_out: *id007
	checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
	checkpoints_dir: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/save
	recoverables:
	counter: *id008
	classifier_enc: *id006
	classifier_out: *id007
	train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
	save_file: results/snrtrain-timedomain-sbpooling-wwhamr-lessstride-stnorm-manyseparators/1234/train_log.txt

	num_separators_per_model: 3
	separator_base_folder: /home/mila/s/subakany/speechbrain_new/recipes/WHAMandWHAMR/separation/results/

	pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
	loadables:
	classifier_enc: !ref <classifier_enc>
	classifier_out: !ref <classifier_out>