{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"},"kaggle":{"accelerator":"none","dataSources":[{"sourceId":70203,"databundleVersionId":8068726,"sourceType":"competition"},{"sourceId":8482692,"sourceType":"datasetVersion","datasetId":5059656},{"sourceId":53383,"sourceType":"modelInstanceVersion","isSourceIdPinned":true,"modelInstanceId":44791}],"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":false}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"import os\nimport glob\nimport librosa\nimport numpy as np\nimport cv2\nimport tensorflow as tf\nimport pandas as pd\nimport pickle\nfrom tensorflow.keras.models import Sequential\nfrom tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, GlobalAveragePooling2D, Input\nfrom tensorflow.keras.optimizers import Adam\nfrom tensorflow.keras.applications import MobileNetV2\nfrom tensorflow.keras.models import Model\n\n# Define the model (assumed structure based on typical CNN for spectrograms)\ndef create_model(input_shape, num_classes):\n    inputs = Input(shape=(224, 224, 3))\n    base_model = MobileNetV2(weights='imagenet', include_top=False, input_tensor=inputs)  # Load pre-trained MobileNetV2\n    x = GlobalAveragePooling2D()(base_model.output)\n    x = Dense(256, activation='relu')(x)\n    outputs = Dense(num_classes, activation='softmax')(x)\n    model = Model(inputs=inputs, outputs=outputs)\n    return model\n\n# Load the model weights\ninput_shape = (224, 224, 3)\nnum_classes = 182\nmodel = create_model(input_shape, num_classes)\nmodel.load_weights('/kaggle/input/mobilenetv2/keras/v1/1/best_model (3).keras')\n\n# Function to preprocess audio to spectrogram using librosa and cv2\ndef preprocess_audio_to_spectrogram(audio_path, img_size=224):\n    audio, sr = librosa.load(audio_path)\n    audio = audio[:sr*5]  # Take a 5-second chunk of audio\n    n_fft = 2048  # Length of the FFT window\n    hop_length = 512  # Number of samples between successive frames\n    n_mels = 128  # Number of Mel bands\n    fmin = 1000  # Min frequency (Hz)\n    fmax = 9000  # Max frequency (Hz)\n\n    # Convert to Mel Spectrogram\n    ms = librosa.feature.melspectrogram(y=audio, sr=sr, n_fft=n_fft, hop_length=hop_length, n_mels=n_mels, fmin=fmin, fmax=fmax)\n    log_ms = librosa.power_to_db(ms, ref=np.max)\n    \n    # Normalize log_ms to be between 0 and 1\n    log_ms_normalized = (log_ms - log_ms.min()) / (log_ms.max() - log_ms.min())\n    \n    # Convert log_ms_normalized to 8-bit unsigned integer format\n    log_ms_normalized_uint8 = (log_ms_normalized * 255).astype(np.uint8)\n    \n    # Convert single-channel image to three-channel RGB image\n    log_ms_rgb = cv2.cvtColor(log_ms_normalized_uint8, cv2.COLOR_GRAY2RGB)\n    \n    # Resize to the target image size for the model\n    log_ms_rgb_resized = cv2.resize(log_ms_rgb, (img_size, img_size))\n    \n    # Normalize the image array\n    log_ms_rgb_resized = log_ms_rgb_resized / 255.0\n    \n    return log_ms_rgb_resized\n\n# Load class names from the pkl file\nwith open('/kaggle/input/class-names/class_names.pkl', 'rb') as f:\n    class_names = pickle.load(f)\n\n# Step 1: Read .ogg files from the test folder\ntest_folder = '/kaggle/input/birdclef-2024/unlabeled_soundscapes'\nogg_files = glob.glob(os.path.join(test_folder, '*.ogg'))\n\n# Step 2: Use the trained model to make predictions directly\npredictions = {}\nfor ogg_file in ogg_files:\n    spectrogram = preprocess_audio_to_spectrogram(ogg_file)\n    spectrogram = np.expand_dims(spectrogram, axis=0)  # Create batch axis\n    prediction = model.predict(spectrogram)\n    predictions[os.path.basename(ogg_file)] = prediction\n\n# Step 3: Create a CSV file with predictions\noutput_file = '/kaggle/working/submission.csv'\ncolumns = ['row_id'] + class_names\ndata = []\n\nfor filename, prediction in predictions.items():\n    row_id = f\"soundscape_{filename.replace('.ogg', '')}\"\n    row = [row_id] + prediction.tolist()[0]\n    data.append(row)\n\ndf = pd.DataFrame(data, columns=columns)\ndf.to_csv(output_file, index=False)\n\nprint(f'Predictions saved to {output_file}')\n","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true},"execution_count":null,"outputs":[]}]}