# ############################################################################
# Model: ECAPA-TDNN for Accent Identification
# ############################################################################

# Pretrain folder (HuggingFace)
pretrained_path: Jzuluaga/accent-id-commonaccent_ecapa

# Feature parameters
n_mels: 80

# Output parameters
n_languages: 16 # Possible languages in the dataset
emb_dim: 192 # dimensionality of the embeddings

# Model params
compute_features: !new:speechbrain.lobes.features.Fbank
    n_mels: !ref <n_mels>

mean_var_norm: !new:speechbrain.processing.features.InputNormalization
    norm_type: sentence
    std_norm: False

# Embedding Model
embedding_model: !new:speechbrain.lobes.models.ECAPA_TDNN.ECAPA_TDNN
    input_size: !ref <n_mels>
    activation: !name:torch.nn.LeakyReLU
    channels: [1024, 1024, 1024, 1024, 3072]
    kernel_sizes: [5, 3, 3, 3, 1]
    dilations: [1, 2, 3, 4, 1]
    attention_channels: 128
    lin_neurons: !ref <emb_dim>

# Classifier based on cosine distance
classifier: !new:speechbrain.lobes.models.ECAPA_TDNN.Classifier
    input_size: !ref <emb_dim>
    out_neurons: !ref <n_languages>

modules:
    compute_features: !ref <compute_features>
    mean_var_norm: !ref <mean_var_norm>
    embedding_model: !ref <embedding_model>
    classifier: !ref <classifier>

label_encoder: !new:speechbrain.dataio.encoder.CategoricalEncoder

pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
    loadables:
        embedding_model: !ref <embedding_model>
        classifier: !ref <classifier>
        label_encoder: !ref <label_encoder>
    paths:
        embedding_model: !ref <pretrained_path>/embedding_model.ckpt
        classifier: !ref <pretrained_path>/classifier.ckpt
        label_encoder: !ref <pretrained_path>/accent_encoder.txt