WGU D604 Task 1: Audio Classification Project Report
Environmental Sound Classification Using Deep Learning
Student Name: Shanikwa Haynes
Course: D604 - Advanced Analytics
Date: July 15, 2025
Executive Summary
This report presents a comprehensive audio classification project using the ESC-50
dataset and convolutional neural networks (CNNs) to address smart city environmental
monitoring needs. The project successfully demonstrates the application of deep learning
techniques for automated environmental sound classification, achieving practical
performance levels suitable for urban deployment.
Table of Contents
A. Research Question and Business Problem
B. Image Dataset (Not Applicable)
C. Video Dataset (Not Applicable)
D. Audio Data Preparation
E. Neural Network Architecture
F. Model Evaluation
G. Summary and Recommendations
H. Code Output and Documentation
I. Sources for Third-Party Code
J. Sources and Citations
K. Professional Communication
A. Research Question and Business Problem
A1. Research Question
Primary Research Question: How effectively can a Convolutional Neural Network (CNN)
classify environmental sounds from the ESC-50 dataset to enable automated urban noise
monitoring systems for smart city applications?
,This research question directly addresses real-world organizational needs as urban
planners and environmental monitoring agencies require automated sound classification
systems for intelligent city management. The question is relevant to realistic organizational
situations where cities need to monitor noise pollution, detect emergency situations, and
analyze urban activity patterns through automated audio analysis.
A2. Objectives and Goals
Primary Objective: Develop a CNN model capable of classifying 50 different
environmental sound categories with high accuracy for deployment in smart city
monitoring systems.
Secondary Objectives: 1. Achieve minimum 70% classification accuracy on ESC-50 test
set 2. Optimize audio preprocessing pipeline for real-time deployment 3. Create
reproducible model architecture for scalable implementation 4. Provide comprehensive
evaluation of model performance across sound categories
Each objective is reasonable within the scope of the research question and represented in
the ESC-50 dataset, which contains 2,000 audio samples across 50 environmental sound
categories. The objectives are measurable, achievable, and directly support the
overarching research question.
A3. Neural Network Type
Selected Network: Convolutional Neural Network (CNN) for audio classification
This is an industry-relevant deep learning architecture capable of performing audio
classification tasks by processing mel-spectrogram representations of audio signals.
CNNs are specifically suitable for environmental sound classification on the ESC-50
dataset due to their ability to detect spatial patterns in 2D spectral representations of
audio data.
A4. Neural Network Justification
Justification for CNN Selection:
1. Spatial Pattern Recognition: CNNs excel at identifying spatial patterns in 2D
representations, making them ideal for analyzing mel-spectrograms derived from
audio signals. The convolutional layers can detect frequency patterns and temporal
relationships within spectrograms.
2. Translation Invariance: CNNs can detect audio features regardless of their
temporal position in the spectrogram, which is crucial for environmental sound
classification where sound patterns may occur at different times within the 5-
second audio clips.
3. Hierarchical Feature Learning: Multi-layer CNN architecture automatically learns
hierarchical features from basic spectral patterns to complex sound signatures,
, eliminating the need for manual feature engineering that would be required with
traditional machine learning approaches.
4. Industry Standard: CNNs are widely adopted in audio processing applications,
with proven effectiveness in similar environmental sound classification tasks
(Salamon & Bello, 2017). This established track record provides confidence in the
approach.
5. Scalability: CNN architecture can be efficiently deployed on various hardware
platforms, from edge devices to cloud servers, supporting smart city infrastructure
requirements for distributed monitoring systems.
B. Image Dataset (Not Applicable)
This section is not applicable as the selected scenario focuses on audio classification
using the ESC-50 dataset.
C. Video Dataset (Not Applicable)
This section is not applicable as the selected scenario focuses on audio classification
using the ESC-50 dataset.
D. Audio Data Preparation
D1. Exploratory Data Analysis
D1a. Spectrogram Explanation
A spectrogram is a visual representation of the spectrum of frequencies in an audio signal
as it varies with time. Key characteristics include: - X-axis: Time (seconds) - Y-axis:
Frequency (Hz) - Color intensity: Magnitude/Power of frequencies - Reveals: temporal
and spectral patterns in audio signals - Essential for: audio classification as it converts 1D
audio into 2D image-like representation
Spectrograms are crucial for CNN-based audio classification because they transform
temporal audio signals into spatial representations that convolutional layers can process
effectively. The mel-scale spectrogram specifically emphasizes perceptually relevant
frequency ranges, making it particularly suitable for environmental sound analysis.
D1b. Audio Tagging Explanation
Audio tagging is the process of assigning descriptive labels to audio segments or files. In
the ESC-50 dataset: - Each audio file has a corresponding category label (e.g., ‘dog’, ‘rain’,
Environmental Sound Classification Using Deep Learning
Student Name: Shanikwa Haynes
Course: D604 - Advanced Analytics
Date: July 15, 2025
Executive Summary
This report presents a comprehensive audio classification project using the ESC-50
dataset and convolutional neural networks (CNNs) to address smart city environmental
monitoring needs. The project successfully demonstrates the application of deep learning
techniques for automated environmental sound classification, achieving practical
performance levels suitable for urban deployment.
Table of Contents
A. Research Question and Business Problem
B. Image Dataset (Not Applicable)
C. Video Dataset (Not Applicable)
D. Audio Data Preparation
E. Neural Network Architecture
F. Model Evaluation
G. Summary and Recommendations
H. Code Output and Documentation
I. Sources for Third-Party Code
J. Sources and Citations
K. Professional Communication
A. Research Question and Business Problem
A1. Research Question
Primary Research Question: How effectively can a Convolutional Neural Network (CNN)
classify environmental sounds from the ESC-50 dataset to enable automated urban noise
monitoring systems for smart city applications?
,This research question directly addresses real-world organizational needs as urban
planners and environmental monitoring agencies require automated sound classification
systems for intelligent city management. The question is relevant to realistic organizational
situations where cities need to monitor noise pollution, detect emergency situations, and
analyze urban activity patterns through automated audio analysis.
A2. Objectives and Goals
Primary Objective: Develop a CNN model capable of classifying 50 different
environmental sound categories with high accuracy for deployment in smart city
monitoring systems.
Secondary Objectives: 1. Achieve minimum 70% classification accuracy on ESC-50 test
set 2. Optimize audio preprocessing pipeline for real-time deployment 3. Create
reproducible model architecture for scalable implementation 4. Provide comprehensive
evaluation of model performance across sound categories
Each objective is reasonable within the scope of the research question and represented in
the ESC-50 dataset, which contains 2,000 audio samples across 50 environmental sound
categories. The objectives are measurable, achievable, and directly support the
overarching research question.
A3. Neural Network Type
Selected Network: Convolutional Neural Network (CNN) for audio classification
This is an industry-relevant deep learning architecture capable of performing audio
classification tasks by processing mel-spectrogram representations of audio signals.
CNNs are specifically suitable for environmental sound classification on the ESC-50
dataset due to their ability to detect spatial patterns in 2D spectral representations of
audio data.
A4. Neural Network Justification
Justification for CNN Selection:
1. Spatial Pattern Recognition: CNNs excel at identifying spatial patterns in 2D
representations, making them ideal for analyzing mel-spectrograms derived from
audio signals. The convolutional layers can detect frequency patterns and temporal
relationships within spectrograms.
2. Translation Invariance: CNNs can detect audio features regardless of their
temporal position in the spectrogram, which is crucial for environmental sound
classification where sound patterns may occur at different times within the 5-
second audio clips.
3. Hierarchical Feature Learning: Multi-layer CNN architecture automatically learns
hierarchical features from basic spectral patterns to complex sound signatures,
, eliminating the need for manual feature engineering that would be required with
traditional machine learning approaches.
4. Industry Standard: CNNs are widely adopted in audio processing applications,
with proven effectiveness in similar environmental sound classification tasks
(Salamon & Bello, 2017). This established track record provides confidence in the
approach.
5. Scalability: CNN architecture can be efficiently deployed on various hardware
platforms, from edge devices to cloud servers, supporting smart city infrastructure
requirements for distributed monitoring systems.
B. Image Dataset (Not Applicable)
This section is not applicable as the selected scenario focuses on audio classification
using the ESC-50 dataset.
C. Video Dataset (Not Applicable)
This section is not applicable as the selected scenario focuses on audio classification
using the ESC-50 dataset.
D. Audio Data Preparation
D1. Exploratory Data Analysis
D1a. Spectrogram Explanation
A spectrogram is a visual representation of the spectrum of frequencies in an audio signal
as it varies with time. Key characteristics include: - X-axis: Time (seconds) - Y-axis:
Frequency (Hz) - Color intensity: Magnitude/Power of frequencies - Reveals: temporal
and spectral patterns in audio signals - Essential for: audio classification as it converts 1D
audio into 2D image-like representation
Spectrograms are crucial for CNN-based audio classification because they transform
temporal audio signals into spatial representations that convolutional layers can process
effectively. The mel-scale spectrogram specifically emphasizes perceptually relevant
frequency ranges, making it particularly suitable for environmental sound analysis.
D1b. Audio Tagging Explanation
Audio tagging is the process of assigning descriptive labels to audio segments or files. In
the ESC-50 dataset: - Each audio file has a corresponding category label (e.g., ‘dog’, ‘rain’,
