examples/speaker_tasks/recognition/conf/SpeakerNet_verification_3x2x256.yaml

name: &name "SpeakerNet"
sample_rate: &sample_rate 16000
repeat: &rep 2
dropout: &drop 0.5
separable: &separable True
n_filters: &n_filters 256

model:
  train_ds:
    manifest_filepath: ???
    sample_rate: 16000
    labels: null
    batch_size: 64
    shuffle: True
    is_tarred: False
    tarred_audio_filepaths: null
    tarred_shard_strategy: "scatter"

  validation_ds:
    manifest_filepath: ???
    sample_rate: 16000
    labels: null
    batch_size: 128
    shuffle: False

  preprocessor:
    _target_: nemo.collections.asr.modules.AudioToMelSpectrogramPreprocessor
    normalize: "per_feature"
    window_size: 0.02
    sample_rate: *sample_rate
    window_stride: 0.01
    window: "hann"
    features: &n_mels 64
    n_fft: 512
    frame_splicing: 1
    dither: 0.00001

  encoder:
    _target_: nemo.collections.asr.modules.ConvASREncoder
    feat_in: *n_mels
    activation: relu
    conv_mask: true

    jasper:
      -   filters: *n_filters
          repeat: 1
          kernel: [3]
          stride: [1]
          dilation: [1]
          dropout: *drop
          residual: true
          separable: *separable

      -   filters: *n_filters
          repeat: *rep
          kernel: [7]
          stride: [1]
          dilation: [1]
          dropout: *drop
          residual: true
          separable: *separable

      -   filters: *n_filters
          repeat: *rep
          kernel: [11]
          stride: [1]
          dilation: [1]
          dropout: *drop
          residual: true
          separable: *separable

      -   filters: *n_filters
          repeat: *rep
          kernel: [15]
          stride: [1]
          dilation: [1]
          dropout: *drop
          residual: true
          separable: *separable

      -   filters: &enc_feat_out 1500
          repeat: 1
          kernel: [1]
          stride: [1]
          dilation: [1]
          dropout: 0.0
          residual: false
          separable: *separable

  decoder:
    _target_: nemo.collections.asr.modules.SpeakerDecoder
    feat_in: *enc_feat_out
    num_classes: 7205
    pool_mode: 'xvector'
    emb_sizes: 256

  loss:
    _target_: nemo.collections.asr.losses.angularloss.AngularSoftmaxLoss # you could also use cross-entrophy loss
    scale: 30
    margin: 0.2

  optim:
    name: sgd
    lr: .006
    weight_decay: 0.001
    momentum: 0.9

    # scheduler setup
    sched:
      name: CosineAnnealing
      warmup_ratio: 0.1
      min_lr: 0.0

trainer:
  devices: 1 # number of gpus
  max_epochs: 200
  max_steps: -1 # computed at runtime if not set
  num_nodes: 1
  accelerator: gpu
  strategy: ddp
  accumulate_grad_batches: 1
  deterministic: True
  enable_checkpointing: False
  logger: False
  log_every_n_steps: 1  # Interval of logging.
  val_check_interval: 1.0  # Set to 0.25 to check 4 times per epoch, or an int for number of iterations
  gradient_clip_val: 1.0

exp_manager:
  exp_dir: null
  name: *name
  create_tensorboard_logger: True
  create_checkpoint_callback: True