Upload 3 files

__init__.py

@@ -0,0 +1,2 @@
+from .configuration_emotion_classifier import EmotionClassifierConfig
+from .modeling_emotion_classifier import EmotionClassifierHuBERT

configuration_emotion_classifier.py

@@ -0,0 +1,14 @@
+from transformers import PretrainedConfig
+
+class EmotionClassifierConfig(PretrainedConfig):
+    model_type = "emotion_classifier"
+
+    def __init__(
+        self,
+        hidden_size=128,
+        num_classes=6,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.hidden_size = hidden_size
+        self.num_classes = num_classes

modeling_emotion_classifier.py

@@ -0,0 +1,29 @@
+from transformers import PreTrainedModel, HubertModel
+import torch.nn as nn
+import torch
+from .configuration_emotion_classifier import EmotionClassifierConfig
+
+
+class EmotionClassifierHuBERT(PreTrainedModel):
+    config_class = EmotionClassifierConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.hubert = HubertModel.from_pretrained("facebook/hubert-large-ls960-ft")
+        self.conv1 = nn.Conv1d(in_channels=1024, out_channels=512, kernel_size=3, padding=1)
+        self.conv2 = nn.Conv1d(in_channels=512, out_channels=256, kernel_size=3, padding=1)
+        self.transformer_encoder = nn.TransformerEncoderLayer(d_model=256, nhead=8)
+        self.bilstm = nn.LSTM(input_size=256, hidden_size=config.hidden_size, num_layers=2, batch_first=True, bidirectional=True)
+        self.fc = nn.Linear(config.hidden_size * 2, config.num_classes)  # * 2 for bidirectional
+
+    def forward(self, x):
+        with torch.no_grad():
+            features = self.hubert(x).last_hidden_state
+        features = features.transpose(1, 2)
+        x = torch.relu(self.conv1(features))
+        x = torch.relu(self.conv2(x))
+        x = x.transpose(1, 2)
+        x = self.transformer_encoder(x)
+        x, _ = self.bilstm(x)
+        x = self.fc(x[:, -1, :])
+        return x