{
  "model_name": "early-fusion-audio-text-emotion-recognition",
  "task": "multimodal_emotion_recognition",
  "dataset": "MELD",

  "device": "cuda",
  "seed": 42,

  "audio": {
    "sampling_rate": 16000,
    "max_duration_sec": 6.0,
    "encoder": "facebook/wav2vec2-base",
    "embedding_dim": 768
  },

  "text": {
    "tokenizer": "bert-base-uncased",
    "encoder": "bert-base-uncased",
    "max_sequence_length": 128,
    "embedding_dim": 768,
    "use_cls_token": true
  },

  "temporal_pooling": {
    "enabled": true,
    "applied_on": "audio_embeddings",
    "pooling_type": "mean+std",
    "time_dimension": 1
  },

  "fusion": {
    "type": "early_fusion",
    "method": "concatenation",
    "fusion_input_dim": 1536
  },

  "classifier": {
    "hidden_dim": 512,
    "num_classes": 7,
    "dropout": 0.3,
    "activation": "relu"
  },

  "training": {
    "epochs": 30,
    "batch_size": 32,
    "learning_rate":  1e-4,
    "optimizer": "Adam",
    "weight_decay": 1e-4,
    "loss_function": "CrossEntropyLoss",
    "use_class_weights": true
  },

  "evaluation": {
    "metrics": ["accuracy", "macro_f1", "weighted_f1"],
    "validation_strategy": "speaker-independent"
  },

  "output": {
    "save_model": true,
    "save_path": "checkpoints/early_fusion_model.pt"
  }
}