Noise Reduction in Audio Signal Using Digital Filtering

Noise Reduction in Audio Signal Using Digital Filtering

Modern audio recordings often suffer from degradation due to unwanted electronic interference. The objective of this project is to enhance a noisy audio signal in which the desired voice signal is contaminated by high-frequency electronic noise.

Theory and Background

• Voice Frequency Range: Typical speech signals occupy the 300 Hz – 3400 Hz range. Because most important speech information exists below 4 kHz, higher frequencies can be filtered without significantly affecting clarity.
• Interference Noise: The noise presents as a high-frequency tone around 6000 Hz, commonly caused by electronic interference, hardware coupling, or power supply noise.
• Notch Filtering: A Notch Filter (Band-Stop Filter) removes a very narrow frequency band while allowing the remaining frequencies to pass through with minimal distortion.

Filter Parameters

Results & Conclusion

Spectral analysis confirmed the presence of a strong peak at 6000 Hz before filtering. After applying the zero-phase IIR notch filter, the peak at 6 kHz was significantly reduced, confirming successful noise suppression. The difference spectrum effectively isolated the removed noise component.

Conclusion: The digital notch filter effectively removed the 6 kHz interference noise from the audio signal while preserving the important speech components. This experiment demonstrates the effectiveness of frequency-selective digital filtering for improving audio quality.

MATLAB Implementation Code

%% FINAL NOISE REMOVAL: NOTCH FILTER CLEANUP
clear; close all; clc;
fprintf('=== FINAL NOISE REMOVAL: NOTCH FILTER CLEANUP ===\n\n');

%% 1. DEFINE PARAMETERS AND LOAD NOISY DATA
noisy_file = 'amir_with_noise.wav';
NOISE_FREQ = 6000; 
[noisy_voice, Fs] = audioread(noisy_file);

%% 2. DESIGN AND APPLY THE SINGLE NOTCH FILTER
Wo = NOISE_FREQ / (Fs/2); % Normalize frequency
Q = 60; % Quality factor
[b, a] = iirnotch(Wo, Wo/Q); % Generate filter coefficients
cleaned_voice = filtfilt(b, a, noisy_voice); % Zero-phase filtering

%% 3. VISUALIZATION AND COMPARISON
N = min(4096, length(noisy_voice));
freqs = Fs*(0:N/2)/N;
difference_signal = noisy_voice - cleaned_voice;

% Calculate the spectrums
Y_noisy = fft(noisy_voice, N);
P_noisy = abs(Y_noisy(1:N/2+1)/N);
Y_cleaned = fft(cleaned_voice, N);
P_cleaned = abs(Y_cleaned(1:N/2+1)/N);
Y_diff = fft(difference_signal, N);
P_diff = abs(Y_diff(1:N/2+1)/N);

% Plot configuration omitted for brevity (Refer to source file for plotting logic)

%% 4. SAVE AND PLAY CLEANED AUDIO
[~, name, ext] = fileparts(noisy_file);
output_file = [name '_CLEANED_FINAL' ext];
audiowrite(output_file, cleaned_voice, Fs);

fprintf('\nSaved cleaned audio as: %s\n', output_file);
soundsc(cleaned_voice, Fs); % Play cleaned audio