Urban Sound classification using CNN

Dataset

• Urban sound 8K dataset

Data analysis

• Analysis of dataset such as recording lengths, class distribution or sampling frequencies.
• Listening to class samples.
• Visualization of data as waveform, STFT, MEL-STFT, MFCC.
• Jupyter notebook
• Google colab

CNN Training

• Trying to classify based on MelSpectogram.
• Data augmented with time shifting, time masking and freq masking.
• Model is saved after each epoch.
• Confusion matrix to peak at model behavior.
• Jupyter notebook
• Google colab

Environment

• Python 3.8 (recommend to create virtual env)

    python38 -m venv .venv
    source .venv/Scripts/Activate
  

• All necessary dependencies in requirements.txt

    pip install -r requirements.txt