FIR Filter for Audio Noise Reduction in Python

Digital Signal Processing Project
Title: Implementation of Finite Impulse Response Filter (FIR) for Noise Reduction in Audio Signals

---

Objective

To reduce unwanted noise such as fan hum and white noise from .wav audio recordings using:

• A Band-pass FIR (Finite Impulse Response) Filter
• An Adaptive Wiener Filter

---

Project Structure

project-root/
├── resource/       # Contains input and output WAV files
├── main.py         # Implements FIR filter using SciPy
├── we.py           # Applies Wiener filter
└── README.md       # This file

---

Methodology

1. Audio Data Acquisition

Audio was recorded in a noisy indoor environment and stored in .wav format for processing.

from scipy.io import wavfile
fs, audio = wavfile.read("resource/output.wav")

---

2. FIR Filter Design

A band-pass FIR filter was created using scipy.signal.firwin to pass frequencies between 1000 Hz and 5000 Hz:

from scipy.signal import firwin

fir_coeff = firwin(numtaps=301, cutoff=[1000, 5000], pass_zero=False)

• Cutoff Frequencies: 1000 Hz – 5000 Hz
• Filter Length: 301 taps (longer filter = sharper cutoff)

---

3. FIR Filter Application

The filter was applied using digital convolution (lfilter) to suppress low and high-frequency noise:

from scipy.signal import lfilter

filtered_audio = lfilter(fir_coeff, 1.0, audio)

---

4. Wiener Filtering

An adaptive Wiener filter was used to further reduce remaining white noise:

from scipy.signal import wiener

filtered_audio = wiener(filtered_audio, mysize=20)

• Window size (mysize): 20 (balance between noise smoothing and signal detail)

---

Results and Observations

• Fan hum (low-frequency noise) and white noise (high-frequency hiss) were significantly reduced.

FIR Filter:

• Sharp band-pass between 1000–5000 Hz.
• Reduced low-frequency rumble and high-frequency noise outside the speech range.

Wiener Filter:

• Adaptively smoothed the residual noise.
• Enhanced clarity by reducing hiss while preserving important signal features.

---

Conclusion

This project successfully demonstrated the practical use of classical DSP techniques — FIR filtering and Wiener filtering — for audio noise reduction using Python.

The combined filtering approach significantly improved audio clarity, making it useful for applications like speech enhancement, telecommunication preprocessing, and audio restoration.

---

References

• Oppenheim & Schafer — Discrete-Time Signal Processing
• R.G. Lyons — Understanding Digital Signal Processing
• SciPy Documentation: scipy.signal.firwin
• Stanford EE264 Lecture on Wiener Filtering
• MIT OCW — Signals, Systems and Inference (Chapter on Wiener Filtering)

---