open_source.yaml

# ############################################################################
# Model: StealthyIMU
# Authors:  Ke Sun, Chunyu Xia 2022
# ############################################################################

# Seed needs to be set at top of yaml, before objects with parameters are made
seed: 1235
__set_seed: !apply:torch.manual_seed [!ref <seed>]
output_folder: !ref results/BPE51_all_opensource_test/<seed>
save_folder: !ref <output_folder>/save
train_log: !ref <output_folder>/train_log.txt

# Data files
file_name: metadata/stealthyIMU_all_relative.csv
data_folder: /data/SoK/StealthyIMU_organize/
train_splits: ["train_synthetic", "train_real"]
csv_train: !ref <output_folder>/train-type=direct.csv
csv_valid: !ref <output_folder>/valid-type=direct.csv
csv_test: !ref <output_folder>/test-type=direct.csv
tokenizer_file: !ref pretrain/51_unigram.model
skip_prep: False

# Training parameters
number_of_epochs: 20
batch_size: 8
lr: 0.0003
token_type: unigram # ["unigram", "bpe", "char"]
sorting: random
ckpt_interval_minutes: 15 # save checkpoint every N min

# Model parameters
sample_rate: 500
n_fft: 80
n_win_length: 80
n_hop_length: 20
n_feature: 31
emb_size: 64
dec_neurons: 256
output_neurons: 51 # index(eos/bos) = 0
ASR_encoder_dim: 256

# Encoding parameters
activation: !name:torch.nn.LeakyReLU
dropout: 0.15
cnn_blocks: 2
cnn_channels: (64, 128)
inter_layer_pooling_size: (2, 2)
cnn_kernelsize: (3, 3)
time_pooling_size: 2
rnn_class: !name:speechbrain.nnet.RNN.LSTM
rnn_layers: 4
rnn_neurons: 256
rnn_bidirectional: True
dnn_blocks: 2
dnn_neurons: 256

# Decoding parameters
bos_index: 0
eos_index: 0
min_decode_ratio: 0.0
max_decode_ratio: 10.0
slu_beam_size: 80
eos_threshold: 1.5
temperature: 1.25

dataloader_opts:
    batch_size: !ref <batch_size>
    shuffle: True

epoch_counter: !new:speechbrain.utils.epoch_loop.EpochCounter
    limit: !ref <number_of_epochs>

normalize: !new:speechbrain.processing.features.InputNormalization
   norm_type: global

compute_features: !new:feature.AccSpec
   sample_rate: !ref <sample_rate>
   win_length: !ref <n_win_length>
   hop_length: !ref <n_hop_length>
   n_fft: !ref <n_fft>

enc: !new:speechbrain.lobes.models.CRDNN.CRDNN
    input_shape: [null, null, !ref <n_feature>]
    activation: !ref <activation>
    dropout: !ref <dropout>
    cnn_blocks: !ref <cnn_blocks>
    cnn_channels: !ref <cnn_channels>
    cnn_kernelsize: !ref <cnn_kernelsize>
    inter_layer_pooling_size: !ref <inter_layer_pooling_size>
    time_pooling: True
    using_2d_pooling: False
    time_pooling_size: !ref <time_pooling_size>
    rnn_class: !ref <rnn_class>
    rnn_layers: !ref <rnn_layers>
    rnn_neurons: !ref <rnn_neurons>
    rnn_bidirectional: !ref <rnn_bidirectional>
    rnn_re_init: True
    dnn_blocks: !ref <dnn_blocks>
    dnn_neurons: !ref <dnn_neurons>

output_emb: !new:speechbrain.nnet.embedding.Embedding
    num_embeddings: !ref <output_neurons>
    embedding_dim: !ref <emb_size>

dec: !new:speechbrain.nnet.RNN.AttentionalRNNDecoder
    enc_dim: !ref <dnn_neurons>
    input_size: !ref <emb_size>
    rnn_type: gru
    attn_type: keyvalue
    hidden_size: !ref <dec_neurons>
    attn_dim: 256
    num_layers: 3
    scaling: 1.0
    dropout: 0.0

seq_lin: !new:speechbrain.nnet.linear.Linear
    input_size: !ref <dec_neurons>
    n_neurons: !ref <output_neurons>

env_corrupt: !new:speechbrain.lobes.augment.EnvCorrupt
    babble_prob: 0.0
    reverb_prob: 0.0
    noise_prob: 1.0
    noise_snr_low: 0
    noise_snr_high: 15

modules:
    enc: !ref <enc>
    output_emb: !ref <output_emb>
    dec: !ref <dec>
    seq_lin: !ref <seq_lin>
    env_corrupt: !ref <env_corrupt>

model: !new:torch.nn.ModuleList
    - [!ref <enc>, !ref <output_emb>,
       !ref <dec>, !ref <seq_lin>]


tokenizer: !new:sentencepiece.SentencePieceProcessor

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
    collect_in: !ref <save_folder>/SLURM_tokenizer
    loadables:
        tokenizer: !ref <tokenizer>
    paths:
        tokenizer: !ref <tokenizer_file>

beam_searcher: !new:speechbrain.decoders.S2SRNNBeamSearcher
    embedding: !ref <output_emb>
    decoder: !ref <dec>
    linear: !ref <seq_lin>
    bos_index: !ref <bos_index>
    eos_index: !ref <eos_index>
    min_decode_ratio: !ref <min_decode_ratio>
    max_decode_ratio: !ref <max_decode_ratio>
    beam_size: !ref <slu_beam_size>
    eos_threshold: !ref <eos_threshold>
    temperature: !ref <temperature>
    using_max_attn_shift: False
    max_attn_shift: 30
    coverage_penalty: 0.

opt_class: !name:torch.optim.Adam
    lr: !ref <lr>

lr_annealing: !new:speechbrain.nnet.schedulers.NewBobScheduler
    initial_value: !ref <lr>
    improvement_threshold: 0.0025
    annealing_factor: 0.8
    patient: 0

checkpointer: !new:speechbrain.utils.checkpoints.Checkpointer
    checkpoints_dir: !ref <save_folder>
    recoverables:
        model: !ref <model>
        scheduler: !ref <lr_annealing>
        counter: !ref <epoch_counter>

log_softmax: !new:speechbrain.nnet.activations.Softmax
    apply_log: True

seq_cost: !name:speechbrain.nnet.losses.nll_loss
    label_smoothing: 0.1

train_logger: !new:speechbrain.utils.train_logger.FileTrainLogger
    save_file: !ref <train_log>

error_rate_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats

cer_computer: !name:speechbrain.utils.metric_stats.ErrorRateStats
    split_tokens: True