“Noise Filtering in Audio Signals Using Fourier Transform”

Project Report

Submitted in Partial Fulfillment for the Degree of Master of Science

(Mathematics)

Under the Supervision of Submitted By

Name

Roll No. :

CENTRE OF DISTANCE & ONLINE EDUCATION VIVEKANAND

GLOBAL UNIVERSITY, JAIPUR

(2023-25)
 CERTIFICATE

Certified that this Project Report entitled “Noise Filtering of Audio Signals using Fourier
Transform” is a record of Project done independently by ……………. under my guidance and
that it has not been previously submitted for the award of any degree, fellowship or associate-ship
to any institute or organization.

(Guide name)
Guide designation
Organization
 DECLARATION

I hereby declare that this Project Report entitled ”Noise Filtering in Audio Signals using Fourier
Transform” is a bonafide record of work done by me during the course of [Link]. Mathematics and
that it has not been previously submitted for the award for any degree/diploma, associate ship,
fellowship or other similar title of any other institute/organization.

Name

Vivekananda Global University, Jaipur

 ACKNOWLEDGEMENT

It is not often in life that we get a chance of appreciating and expressing our feelings in black and
white to thank the people who have been a crucial part of our successes, our accomplishments, and
our being what we are today. I take this opportunity to first of all thank the Faculty at Vivekananda
Global University, especially Dr. Arvind Kumar Singh, Director, CDOE , Dr. Mukesh Agnihotri ,
Program Coordinator and Aafrin Gouri mam, Assistant Professor (Mathematics) Vivekananda
Global University Jaipur for inculcating and instilling the knowledge, learning, will-power,
values and the competitiveness and professionalism required by me as a Post Graduate learner.

I would like to give special thanks to Sh. Prasun & Mrs. Somya Bhardwaj (Mentors of CDOE,
Vivekananda Global University, Jaipur) for their enduring efforts, guidance, patience and
enthusiasm which has given a sense of direction and purposefulness to this Project Report and
ultimately made it a success.

Last but not the least; I would like to thank my family, my parents and friends, for supporting me
spiritually throughout the completion of the Project Report. The errors and inconsistencies remain
my own.

Rahul Namawat

VGU23ONS3MSA0009
 Contents

Abstract

Chapter 1 Introduction

Chapter 2 Mathematical Background

2.1 Fourier Series

2.2 Fourier Transform

2.3 Discrete Fourier Transform

2.4 Fast Fourier Transform

Chapter 3 Problem Modelling

3.1 Mathematical Formulation

3.2 Real World Challenges in Audio Signal Processing

3.3 The Role of Fourier Transform in Noise Filtering

3.4 The Objective of Filtering

Chapter 4 MATLAB based explanation of Filtering

Chapter 5 Results (With Illustrative Plots)

5.1 Time Domain Analysis

5.2 Frequency Domain Analysis

Chapter 6 Conclusion from results

Chapter 7 References
 Abstract

In this project, we explore the application of Fourier Transform in removing noise from audio
signals. Noise filtering is a critical task in audio signal processing, with applications ranging from
telecommunications to hearing aids and media restoration. The project provides a theoretical
overview of Fourier analysis, followed by a discussion on signal representation in the frequency
domain. A conceptual MATLAB-based method is outlined for identifying and filtering noise
components from an audio waveform using the Fast Fourier Transform (FFT).
1. Introduction
In the modern world, audio communication plays a vital role across various domains, including
telephony, broadcasting, and medical diagnostics. However, audio signals are often corrupted by
noise during acquisition or transmission. Filtering out such noise is essential for accurate
interpretation and clarity.

Fourier Transform provides a powerful tool for analyzing signals in the frequency domain,
enabling the separation of noise from meaningful information. This project examines the
mathematical foundation and practical application of Fourier Transform for audio noise filtering,
particularly focusing on frequency-based identification and removal of unwanted components.

2. Mathematical Background
Fourier analysis is a mathematical technique that allows decomposition of functions into
sinusoidal components. This is particularly useful in signal processing, where signals can be
represented as sums of sine and cosine waves of different frequencies.

2.1 Fourier Series

If a function f(t) is periodic with period T, then it can be represented by a Fourier Series:
f(t) = a₀/2 + Σ [aₙ cos(nω₀t) + bₙ sin(nω₀t)], n = 1 to ∞
where ω₀ = 2π/T is the fundamental angular frequency.
The coefficients aₙ and bₙ are given by :
aₙ = (2/T) ∫₀ᵗ f(t) cos(nω₀t) dt
bₙ = (2/T) ∫₀ᵗ f(t) sin(nω₀t) dt

2.2 Fourier Transform

For non-periodic signals, the Fourier Transform is used to analyze the frequency spectrum:
F(ω) = ∫ f(t) e^(-jωt) dt
The inverse Fourier Transform is used to reconstruct the signal:
f(t) = (1/2π) ∫ F(ω) e^(jωt) dω

2.3 Discrete Fourier Transform (DFT)

In practice, signals are digital and the Discrete Fourier Transform (DFT) is applied. Given a
discrete signal x[n] of length N:
X[k] = Σ x[n] e^(-j2πkn/N), n = 0 to N-1
The Inverse DFT is:
x[n] = (1/N) Σ X[k] e^(j2πkn/N), k = 0 to N-1

2.4 Fast Fourier Transform (FFT)

FFT is an efficient algorithm to compute the DFT. MATLAB provides the 'fft' and 'ifft' functions
to perform these operations quickly, even for large data sets.

3.0 Problem Modeling

In practical environments, audio signals are seldom captured in ideal conditions. They are
frequently contaminated by various types of noise, which obscure the original message or content
within the signal. Noise can be introduced during acquisition, transmission, or playback, and its
impact is particularly significant in fields like telecommunications, biomedical signal processing,
and multimedia.

Types of Noise Affecting Audio Signals

1. Electromagnetic Interference (EMI):

EMI originates from surrounding electrical and electronic devices. Common sources include
motors, transformers, fluorescent lights, switching power supplies, and unshielded cables. This
form of noise typically introduces high-frequency components that distort the integrity of the
original signal.
2. Environmental or Background Noise:
These are ambient sounds present in the environment during signal acquisition. Examples
include:
 Traffic sounds in outdoor recordings.
 Fan or HVAC (heating, ventilation, and air conditioning) noise in indoor settings.
 Crowd noise or human speech in public places.
3. Recording Equipment Imperfections:
Noise introduced by the limitations of recording hardware includes:
 Microphone static or hiss.
 Distortion from over-amplification.
 Quantization noise in analog-to-digital conversion.

These interferences degrade the Signal-to-Noise Ratio (SNR) and reduce the clarity and
usability of the recorded audio.
The task of noise filtering involves identifying and isolating these unwanted components from the
original, meaningful audio signal.

3.1 Mathematical Formulation

The noisy signal y(t) can be expressed as:

y(t)=s(t)+n(t)

Where:

 y(t) is the observed (noisy) signal.

 s(t) is the original (desired) signal.
 n(t) is the noise component.

The goal of noise filtering is to recover the clean signal s(t) by estimating and removing n(t) from
y(t). Using Fourier Transform, this problem can be approached by analyzing the frequency
domain representation of the signal, where noise typically appears at specific frequency ranges.
By applying a filter to eliminate those frequencies, we can reconstruct a cleaner version of the
signal.
3.2 Real-World Challenges in Audio Signal Processing

These interferences significantly degrade the Signal-to-Noise Ratio (SNR), a critical metric in
signal processing that quantifies the proportion of useful signal power relative to the background
noise power. A low SNR indicates that noise levels are comparable to or exceed the strength of
the desired signal, making it difficult to accurately interpret or analyze the content. This
degradation manifests as:

 Reduced clarity in speech or music, making it harder to distinguish between different

sounds.
 Loss of important signal features, such as subtle tonal variations or transient components.
 Increased error rates in digital communication systems and audio recognition algorithms.
 Listener fatigue and reduced intelligibility, especially in assistive listening or
telecommunication applications.

In professional applications—like medical diagnostics (e.g., ECG, EEG), voice-controlled

systems, and broadcast media—maintaining a high SNR is crucial to ensuring that the signal is
both perceptually clear and mathematically analysable. Thus, noise filtering becomes an
essential pre-processing step to enhance signal fidelity and usability.

3.3 The Role of the Fourier Transform in Noise Filtering

The Fourier Transform provides a powerful tool for analyzing signals in the frequency
domain. In the time domain, noise and signal components may overlap and be difficult to
distinguish. However, in the frequency domain:

 Periodic signals often concentrate energy in narrow frequency bands.

 Noise is often spread across a wide range of frequencies or located in specific bands (e.g.,
high-frequency hiss or low-frequency hum).

By transforming y(t)y(t) into the frequency domain:

Y(f)=S(f)+N(f)Y(f) = S(f) + N(f)

We can analyze and selectively attenuate the frequency components corresponding to N(f)N(f),
while preserving S(f)S(f). This is typically done using:

 Low-pass filters to remove high-frequency noise.

 Band-stop filters to eliminate specific known interference bands (e.g., 50/60 Hz power
line noise).
 Spectral subtraction or Wiener filtering for more adaptive techniques.

3.4 Objective of Filtering

The main goal is to reconstruct a clean estimate s^(t)\hat{s}(t) such that:

s^(t)≈s(t)

This process enhances audio quality, improves intelligibility, and facilitates further processing
such as compression, recognition, or transmission.
This image is a scientific-style diagram illustrating the process of noise filtering in audio
signals using Fourier Transform, and it's broken into four key stages:

1. Noisy Audio Signal (Top Panel)

 Shows a time-domain waveform with irregular, high-frequency noise.

 This is what a raw, unfiltered audio signal might look like after recording or transmission.

2. Frequency-Domain Representation

 Using the Fourier Transform, the signal is converted from time to frequency domain.
 The frequency spectrum reveals which parts of the signal contain dominant frequencies.
 A box labeled “Filter” indicates that certain frequencies (likely corresponding to noise) are
being removed.
3. Inverse Fourier Transform

 After filtering, the modified frequency data is transformed back to the time domain using
the Inverse FFT.
 This yields a cleaner signal with reduced high-frequency noise.

4. Cleaned Signal (Bottom Panel)

 Final time-domain signal is visibly smoother and cleaner.

 Most of the noise has been suppressed, while retaining the original waveform’s core shape.

This image effectively communicates the core concept of Fourier-based signal denoising:
Transform → Filter → Inverse Transform → Clean Output.

4.0 MATLAB-based Explanation of Filtering

The noise filtering process in MATLAB involves several key steps, which can be implemented
using built-in functions such as fft, ifft, and audioread. Below is a step-by-step explanation of
how to perform the filtering operation:

1. Load the Audio Signal:

To begin, we need to load the audio signal from a file. MATLAB's audio read function reads an
audio file into a variable. For example, if we have an audio file named "[Link]," we can
load it using the following command:

[y, Fs] = audioread('[Link]');

Here:

 y is the vector containing the audio signal.

 Fs is the sampling rate (number of samples per second).

2. Compute the Fast Fourier Transform (FFT):

The next step is to convert the audio signal from the time domain to the frequency domain using
the Fast Fourier Transform (FFT). The fft function in MATLAB computes the Fourier transform:

Y = fft(y);
This returns the frequency-domain representation of the signal in Y. We can visualize this
transformation by plotting the magnitude of Y:

plot(abs(Y));
title('Frequency Spectrum of the Noisy Signal');
xlabel('Frequency (Hz)');
ylabel('Magnitude');

3. Apply a Frequency Filter:

Once the FFT is computed, we can apply a filter to suppress or remove specific frequency
components associated with noise. For example, we can remove high-frequency noise by setting
the values of Y at certain frequencies to zero. Suppose we want to filter out noise above 500 Hz:

% Define the frequency range to be removed (for example, above 500 Hz)
freq_range = abs(freq) > 500; % freq is the frequency axis corresponding to Y

% Zero out the high-frequency components

Y_filtered = Y;
Y_filtered(freq_range) = 0;

The variable Y_filtered now contains the frequency components of the signal with noise
removed.

4. Inverse FFT (IFFT):

After applying the filter, the filtered signal is returned to the time domain using the inverse FFT
(ifft), which reconstructs the signal from its frequency components:

y_filtered = ifft(Y_filtered);

5. Listen to or Save the Filtered Audio:

After filtering, we can listen to the filtered audio using the sound function, which plays the audio
through the system’s speaker:

sound(real(y_filtered), Fs);

Alternatively, we can save the filtered audio to a new file using the audiowrite function:
audiowrite('filtered_audio.wav', real(y_filtered), Fs);

5.0 Results (with Illustrative Plots)

After applying the noise filtering technique, the filtered audio signal can be analyzed in both the
time domain and the frequency domain to evaluate the effectiveness of the filter.

5.1Time-Domain Analysis:

In the time domain, the original noisy signal can be compared with the filtered signal to assess
the removal of noise. The following plot shows the noisy signal and the filtered signal, with the
noise reduced in the latter.

Original (Noisy) vs Filtered Signal in Time Domain:

figure;
subplot(2, 1, 1);
plot(t, y); % Original noisy signal
title('Original (Noisy) Signal');
xlabel('Time (s)');
ylabel('Amplitude');

subplot(2, 1, 2);
plot(t, real(y_filtered)); % Filtered signal
title('Filtered Signal');
xlabel('Time (s)');
ylabel('Amplitude');
In this plot:

 The original signal (top) exhibits the noise (high-frequency fluctuations).

 The filtered signal (bottom) demonstrates a much cleaner waveform with significantly
reduced noise, preserving the essential features of the original signal.

5.2 Frequency-Domain Analysis:

To further confirm the effectiveness of the filtering, we examine the frequency domain before
and after filtering. The FFT of the noisy signal reveals a broad spectrum of high-frequency noise.
After filtering, these high-frequency components are removed, and the frequency spectrum
becomes cleaner.

Frequency Spectrum Before and After Filtering:

% Frequency spectrum of the original noisy signal

figure;
subplot(2, 1, 1);
plot(abs(Y));
title('Frequency Spectrum of the Noisy Signal');
xlabel('Frequency (Hz)');
ylabel('Magnitude');

% Frequency spectrum of the filtered signal

subplot(2, 1, 2);
 plot(abs(Y_filtered));
title('Frequency Spectrum of the Filtered Signal');
xlabel('Frequency (Hz)');
ylabel('Magnitude');

In this plot:

 The noisy signal (top) has a broad spectrum, with large magnitude peaks at higher
frequencies corresponding to the noise.
 The filtered signal (bottom) shows a reduced spectrum, where high-frequency noise
components are significantly attenuated.

6.0 Conclusion from Results:

From both the time-domain and frequency-domain analysis, it is evident that the filtering process
successfully removed much of the noise from the original signal while preserving the essential
characteristics of the clean signal. The effectiveness of the noise reduction is most noticeable in
the frequency domain, where unwanted high-frequency components are eliminated. This
demonstrates the power of Fourier Transform-based filtering in real-world audio processing
tasks.
6.1 Conclusion

This project demonstrates the application of Fourier Transform for noise filtering in audio
signals. By converting the signal from the time domain to the frequency domain using the Fast
Fourier Transform (FFT), we can identify and isolate the noise components, which typically
appear at specific frequencies. After applying a suitable filter to remove these unwanted
frequencies, the filtered signal can be reconstructed using the inverse FFT (IFFT).

The filtering process effectively reduces high-frequency noise, preserving the integrity of the
original audio signal. This technique is particularly useful in various practical applications, such
as:

 Audio restoration: Cleaning up old or noisy recordings.

 Telecommunication systems: Reducing interference in audio signals.
 Multimedia processing: Enhancing the quality of digital audio in applications such as
voice recognition, streaming, and recording.

In conclusion, Fourier Transform-based noise filtering provides an efficient and reliable method
for improving the quality of audio signals, making it a fundamental tool in digital signal
processing.

7.0 References

1. Oppenheim, A. V., & Schafer, R. W. (2009). Discrete-Time Signal Processing. Pearson.

2. Proakis, J. G., & Manolakis, D. G. (2007). Digital Signal Processing: Principles,
Algorithms, and Applications. Prentice Hall.
3. MATLAB Documentation: [Link]
4. Smith, Steven W. (1997). The Scientist and Engineer's Guide to Digital Signal Processing.
California Technical Publishing.