Birdcall identification (Kaggle Competition)
Analysis, Extraction of spectrogram image information

Introduction

BirdCLEF 2021 is a Kaggle competition[1] that aims to classify bird songs by species. This task is very complex due to the very noisy recordings. We propose to extract the mel-spectrogram from the audio files and use a convolutional layer network to perform the classification. ResNet models seem the most promising although pre-training these networks does not seem to be beneficial. Resnet 34 obtains a test accuracy of 52.48%. This suggests that the task is feasible but that there is always room for improvement.

Repository

In the data folder, there are the input and output subfolders.

In input, there are the files necessary to run the program. In output, it is the result of the transformation of sound data into images used during the classification.

In code, there is the program code. There are two identical files, one of which is executable with Jupyter Notebook and another with Python.

In the images folder, there are three images used in the wiki to visualize the results of each step.

• code

  • birdcall_identification.ipynb
  • birdcall_identification.py

• data

  • input
    • test_soundscapes
      • COL_recording_location.txt
      • COR_recording_location.txt
      • SNE_recording_location.txt
      • SSW_recording_location.txt
      • test_set_recording_dates.csv
    • train_soundscapes
      • 2782_SSW_20170701.ogg
      • 7019_COR_20190904.ogg
      • ...
      • 54955_SSW_20170617.ogg
      • 57610_COR_20190904.ogg
    • train_short_audio
      • acafly
        • XC6671.ogg
        • ...
        • XC600277.ogg
      • ... (not here)
    • sample_submission.csv
    • train_soundscape_labels.csv
    • train_metadata.csv
    • test.csv
  • output
    • train_img
      • acafly
        • 27_0.jpg
        • ...
        • 131_2.jpg
      • ...
      • yetvir
        • 22_0.jpg
        • ...
        • 106_2.jpg
    • test_img
      • acafly
        • 0_0.jpg
        • ...
        • 26_1.jpg
      • ...
      • yetvir
        • 0_0.jpg
        • ...
        • 21_2.jpg

• images

  • bird-discussion.png
  • bird-validation.png
  • bird-experimentation.png