Create configuration_ecapa.py

configuration_ecapa.py

@@ -0,0 +1,65 @@
+from transformers.configuration_utils import PretrainedConfig
+
+
+class EcapaConfig(PretrainedConfig):
+
+    model_type = 'ecapa'
+
+    def __init__(
+        self, 
+        n_mels=80, 
+        sample_rate=16000, 
+        win_length=25, 
+        hop_length=10, 
+        mean_norm=True, 
+        std_norm=False, 
+        norm_type='sentence', 
+        hidden_size=192, 
+        channels=[512, 512, 512, 512, 1536],
+        kernel_sizes=[5, 3, 3, 3, 1],
+        dilations=[1, 2, 3, 4, 1],
+        attention_channels=128,
+        res2net_scale=8,
+        se_channels=128,
+        global_context=True,
+        groups=[1, 1, 1, 1, 1], 
+        num_classes=1251, 
+        loss_fn='aam', 
+        auto_map={
+            "AutoConfig": "configuration_ecapa.EcapaConfig",
+            "AutoModel": "modeling_ecapa.EcapaModel", 
+        }, 
+        initializer_range=0.02,
+        **kwargs
+    ):
+        # Compute features
+        self.n_mels = n_mels
+        self.sample_rate = sample_rate
+        self.win_length = win_length
+        self.hop_length = hop_length
+
+        # Mean variance norm
+        self.mean_norm = mean_norm
+        self.std_norm = std_norm
+        self.norm_type = norm_type
+
+        # Embedding model
+        self.channels = channels
+        self.kernel_sizes = kernel_sizes
+        self.attention_channels = attention_channels
+        self.dilations = dilations
+        self.res2net_scale = res2net_scale
+        self.se_channels = se_channels
+        self.global_context = global_context
+        self.groups = groups
+        self.hidden_size = hidden_size
+
+        # Classifier
+        self.num_classes = num_classes
+        self.loss_fn = loss_fn
+
+        # Others
+        self.auto_map = auto_map
+        self.initializer_range = initializer_range
+
+        super().__init__(**kwargs)