Audio Classification Project Using Deep Learning

A deep learning-based project to classify urban sounds into distinct categories using the UrbanSound8K dataset.

Technologies Used

• Python: Programming language.
• NumPy: For numerical computations.
• Pandas: For handling dataset metadata.
• Librosa: For audio signal processing and feature extraction.
• Matplotlib: For visualizations.
• TensorFlow/Keras: For building and training the CNN model.

Dataset

The dataset used in this project is UrbanSound8K, which contains 8,732 labeled sound excerpts (≤4 seconds) of urban sounds belonging to 10 classes:

• air_conditioner
• car_horn
• children_playing
• dog_bark
• drilling
• engine_idling
• gun_shot
• jackhammer
• siren
• street_music

Metadata File: UrbanSound8K.csv

This file provides detailed information about each audio file in the dataset, including:

• slice_file_name: The audio file name in the format [fsID]-[classID]-[occurrenceID]-[sliceID].wav
  • [fsID]: Freesound ID of the recording.
  • [classID]: Numeric identifier of the sound class (e.g., 0 = air_conditioner).
  • [occurrenceID]: Distinguishes occurrences of the sound in the original recording.
  • [sliceID]: Distinguishes slices from the same occurrence.
• fsID: Freesound ID of the recording.
• start / end: Start and end time of the slice in the original recording.
• salience: A subjective rating of the sound’s prominence (1 = foreground, 2 = background).
• fold: Fold number (1–10) for cross-validation.
• classID: Numeric identifier of the sound class (e.g., 0 = air_conditioner).
• class: Human-readable class name (e.g., "air_conditioner").

Objective

This study aims to classify urban sounds into the aforementioned categories using a Convolutional Neural Network (CNN) model.

Project Files

1. preprocessing.ipynb

   • Performs preprocessing steps such as:
     • Audio file loading.
     • Feature extraction (e.g., Mel-spectrograms).
     • Dataset preparation for model training.

2. model_preparation_and_training.ipynb

   • Builds and trains a CNN model on the processed data.
   • Evaluates model performance using metrics such as loss and accuracy.

Results

Top Three Training Results:

Model	Loss	Accuracy
Model 1	0.4307	86.87%
Model 2	0.3629	89.27%
Model 3	0.4355	88.01%

How to Use

Step 1: Clone the Repository

git clone https://github.com/your-username/audio-classification-project.git
cd audio-classification-project

Step 2: Install Dependencies

Ensure you have Python and the necessary libraries installed. You can install the dependencies using:

pip install numpy pandas librosa matplotlib tensorflow keras

Step 3: Run Preprocessing

Run the preprocessing.ipynb notebook to preprocess the dataset and extract features.

jupyter notebook preprocessing.ipynb

Step 4: Train the Model

Run the model_preparation_and_training.ipynb notebook to train and evaluate the CNN model on the processed data.

jupyter notebook model_preparation_and_training.ipynb