Urban Audio Classification using CNN

This project aims to classify urban sounds from the UrbanSound8K dataset into 10 different classes using Convolutional Neural Networks.

Introduction

Urban sound classification involves categorizing various urban sounds into distinct classes, such as sirens, car horns, street music, and more. This project uses machine learning to automate the classification process based on audio data.

Dataset

The UrbanSound8K dataset is a compilation of urban sound recordings, classified in 10 categories according to the paper "A Dataset and Taxonomy for Urban Sound Research", which proposes a taxonomical categorization to describe different environmental sound types.

The UrbanSound8K dataset contains 8732 labeled sound slices of varying duration up to 4 seconds. The categorization labels being:

• Air Conditioner
• Car Horn
• Children Playing
• Dog bark
• Drilling
• Engine Idling
• Gun Shot
• Jackhammer
• Siren
• Street Music

Note that the dataset comes already organized in 10 validation folds. In the case we want to compare our results with other we should stick with this schema.

• Dataset metadata The included metadata file (audio_data/metadata/UrbanSound8K.csv) provides all the required information about each audio file:

  • slice_file_name: The name of the audio file.
  • fsID: The Freesound ID of the recording from which this excerpt (slice) is taken
  • start: The start time of the slice in the original Freesound recording
  • end: The end time of slice in the original Freesound recording
  • salience: A (subjective) salience rating of the sound. 1 = foreground, 2 = background.
  • fold: The fold number (1-10) to which this file has been allocated.
  • classID: A numeric identifier of the sound class.
  • class: The class label name.

Project Structure

The project directory is organized as follows:

.
├── README.md
├── audio_data.dvc
├── dvc.lock
├── dvc.yaml
├── metrics.csv
├── params.yml
├── requirements.txt
├── audio_data
    ├── audio
        ├── fold1
          ├── .wav
          ├── .wav
        ├── fold2
        ...
        ├── fold10
    ├── metadata
      ├── UrbanSound8K.csv
└── src
    ├── __init__.py
    ├── const.py
    ├── data
    │   ├── __init__.py
    │   └── generator.py
    ├── model
    │   ├── __init__.py
    │   └── train.py
    └── notebook
        └── Urban_Audio_Classifier_.ipynb

Dependencies

Ensure you have the following dependencies installed:

• Python 3.x
• librosa 0.9.2
• pandas 1.3.5
• tensorflow 2.11.0
• tqdm 4.64.1
• dagshub 0.2.4
• dvc 2.10.2
• mlflow 1.26.1
• Installation

Usage

• Clone the repository

git clone https://dagshub.com/nirbarazida/urban-audio-classifier.git
cd urban-audio-classifier

• Install the required packages using pip:

pip install -r requirements.txt

• Training (The training process will be logged using DagsHub and MLflow)

  • Using DVC's Data Pipeline

  MLFLOW_TRACKING_URI=https://dagshub.com/{DAGSHUB_REPO_OWNER}/{DAGSHUB_REPO_NAME}.mlflow \
  MLFLOW_TRACKING_USERNAME={DAGSHUB_USER_NAME} \
  MLFLOW_TRACKING_PASSWORD={DAGSHUB_TOKEN} \
  dvc repro
  

  • Using the train script

  MLFLOW_TRACKING_URI=https://dagshub.com/{DAGSHUB_REPO_OWNER}/{DAGSHUB_REPO_NAME}.mlflow \
  MLFLOW_TRACKING_USERNAME={DAGSHUB_USER_NAME} \
  MLFLOW_TRACKING_PASSWORD={DAGSHUB_TOKEN} \
  python -m src.model.train