audio preprocessing

[sorokivski]

Audio Preprocessing for Bird Call Extraction and Spectrogram Conversion

1. Bird Call Extraction

Goal: 🔸Isolation of the audio fragments from files that contain just the bird calls.

2. Zero Padding for Uniform Length

Goal: 🔸 Ensure all audio clips have a uniform length for model compatibility. implementation: finding the longest call duration and apply zero padding for others to make them same length

why? learning models like CNNs or RNNs, which expect inputs of fixed size

3. Spectrogram Conversion

Goal: 🔸Convert the processed audio clips into spectrograms for model input

Expected result:

🔸spectrograms of bird calls input with zero padding in .npy format in folder /data/preprocessed with exact names as a videofiles from which they was extracted.

[SARTARUN]

• The file has been uploaded to isolate the audio file using two parameters: start_cut and end_cut from the main.py file.

• The padding with the zeros has been created. After creating the spectrogram values, they have been saved in the mentioned path with the name corresponding to the actual file name. They saved in .npy file. The mentioned folder provide the example of the output data. The whole data of .npy needs to be transfer via SSD ;)

• From the file name main.py, all the parameters might be changed.