{
  "model_type": "wav2vec2",
  "architectures": ["Wav2Vec2ForSequenceClassification"],
  "input_channels": 1,
  "conv_layers": [
    {
      "in_channels": 1,
      "out_channels": 32,
      "kernel_size": 3,
      "stride": 1,
      "padding": 1
    },
    {
      "in_channels": 32,
      "out_channels": 64,
      "kernel_size": 3,
      "stride": 1,
      "padding": 1
    }
  ],
  "pooling_layers": [
    {
      "kernel_size": 2,
      "stride": 2
    },
    {
      "kernel_size": 2,
      "stride": 2
    }
  ],
  "fc_layers": [
    {
      "in_features": 1763968,  // Adjust this based on your actual input size
      "out_features": 128
    },
    {
      "in_features": 128,
      "out_features": 1
    }
  ],
  "dropout_rate": 0.25,
  "activation_function": "relu",
  "output_size": 1
}