See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/330521494
Sound Editing using Fourier Transform (Term Paper)
Preprint · January 2019
DOI: 10.13140/RG.2.2.13640.16645
CITATION READS
1 4,838
1 author:
Ashutosh Chapagain
Kathmandu University
2 PUBLICATIONS 3 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
Sound Editing using Fourier Transform View project
Image generation View project
All content following this page was uploaded by Ashutosh Chapagain on 21 January 2019.
The user has requested enhancement of the downloaded file.
Sound Editing using Fourier Transform
Ashutosh Chapagain(08), Krishna Gaire(10) ,Anup Sedhain(43), Krishu Thapa(58), Aazad Dahal(67)
Department Of Computer Science And Engineering, School Of Engineering,
Kathmandu University
Abstract
The ability to reduce a complex continuous waveform is highly desirable in a wide variety
of fields like compression and analysis. This paper visually demonstrates simple methods to edit
unwanted notes (individual note or noise) from a given sound clip with Fourier Transform. Fourier
spectra does an excellent job of identifying the frequency content of individual notes. The Fourier
transform is like a mathematical prism where we feed in a wave and it spits out the ingredients of
that wav. We have discussed the theory aspect of how Discrete Fourier transform can be used to
analyze continuous signals like sound. We have demonstrated a method to reduce unwanted note in
a sound using Fourier Analysis. For this we have used a digital audio editor and recording
application software called Audacity.
1. Introduction
Television, radio, audio speakers generate a considerable amount of sound. During the
generation of these sound, devices might pick up electromagnetic waves which can noise the
original sound. These sound which can be stored in a device and are picked up as electromagnetic
waves can be represented as digital sound. A wide variety of tools are available to the artist working
with digital sound. Sound editors range from open source and free software (Ardour, Audacity) to
the professional-level studio (Ableton Live, Logic Pro, Digital Performer, Nuendo, Pro Tools) and
they all perform a variety of Digital Signal Processing tasks, such as signal normalization, fading
edits, and sample-rate conversion [1].
Figure 1.1: Audacity- An open source sound-editing tool
Fourier Series is a mathematical idea that applies itself to almost everything in our real world. The
Fourier Transform gives us a powerful way of viewing these waveform. Fourier transform is
analogous to separating the mixed paint [2] .If the cake is the waveform, the recipe is the Fourier
transform, a list of ingredients and how to combine them. Its major applications include separating
the signal from the noise in data collected for astronomy, medicine, genetics and chemistry.
The member of Fourier series family which is related to digital signal processing is Discrete Fourier
Transform (DFT) which operates on a sampled time domain which is periodic. Any digital sound, f
is a sum of pure sounds at different frequencies [3].Consider a digital sound x[n] that is periodic
with period N, so that x[n] =x[n+rN] for any integer values of n and r. Such sequence can be
represented by a Fourier series corresponding to a sum of harmonically related complex exponential
sequence. These periodic complex exponents are of the form:
ek [n]=ej(2π/N)kn =ek[n+rN]
where k is an integer, and the Fourier Series representation has the form
x [n] = 1/N ΣkX[k]ej(2π/N)kn .[6]
Fourier Series allows experimentalists to identify sources of sound [7]. In our application, to
eliminate the sources of sound, we have to use Inverse Fourier Transform. In general, the goal of
Inverse Fourier Transform is to take x[n] and use them to reconstruct the original signal.
2. Discussion
The basis steps in applying Discrete Fourier Transform (DFT) to continuous-time signals are
indicated in figure . The antialiasing filter is incorporated to eliminate or minimize the effect of
aliasing when continuous-time signal is converted to sequence. x[n] is multiplied by a window size
w[n], for finite length required in DFT. In analysing music, x[n] is indefinitely long. Therefore a
finite-duration window x[n] is applied to x[n] prior to computation of DFT.
Figure 2.1: Processing steps in the discrete-time Fourier Analysis of a continuous-time signal [6]
The conversion of xc[t] to the sequence of samples x[n] is represented in frequency domain by
periodic replication and frequency normalization i.e.
As indicated, sequence x[n] is multiplied by a finite-duration window w[n], since the input to DFT
must be of finite duration. This produces the finite length sequence v[n]=w[n]*x[n]. The effect in
frequency domain is a periodic convolution, i.e,
Convolution of W(ejw) and X(ejw) will tend to smooth sharp peaks and discontinuities in X(ejw). The
DFT of windowed sequence V[n]=w[n]*x[n] is,
Audacity’s spectrum analyzers are based on this theory. This method is useful to clean up a
recording – if there is a Hiss and pop in sound recordings, it can be cleaned up by doing spectrum
analysis. The unwanted individual notes can be broadly classified to:
1. Removing an unwanted note
2. Removing Noise
1. Removing an Unwanted Note
Suppose we have a digital sound which has an annoying high pitch we’d like to edit. The sound
comes as a function of various intensities over time (blue signals) which compose to form a
composite sound wave (the black wave in the figure below). The Fourier Transform of the sound
wave will show a sharp spike corresponding to amplitude from individual notes. We can filter the
unwanted note by smashing its spike down. We then apply the inverse Fourier Transform to all
notes to form the edited digital sounds.
Figure 2.2: A composite sound signal (black) consisting of multiple pure signals signals (blue) with
different frequencies (red).
2. Removing Noise
Discrete Fourier Transform is also used in denoising the sound. Here small high-frequency
components are discarded. After reconstruction of the original sound waveform, the sound becomes
noise free. [5]. Fourier transforms can also be used to extract signals that are so deeply buried in
noise that the human ear would not notice them. This allows communication with very low power
3. Demonstration
We have used Audacity, which is an open source sound editing application to separate
individual notes from the composed songs. A whistle is recorded from the stereo microphone. If
properly recorded, a whistle will form a sine wave [8].
Figure 3.1: A whistle recorded from microphone (Sine Wave)
Plots are made using a Fast Fourier Transform or FFT. This gives a value for each narrow band of
frequencies that represents how much of those frequencies is present. All the values are then
interpolated to create the graph below. Here "dB" is a ratio. It is not an "absolute" measure, but is a
measure of something compared to another (compared to a reference), where the reference level is
said to be "0 dB".
Figure 3.2: Frequency of waveform of Figure 3.1
Now, audacity will apply Fourier Transform [8] to produce a spectrogram of varying frequencies
between 0 Hz and 8000 Hz as shown in figure below.
Figure 3.3: Sound spectrum obtained by Fourier Transform
The default frequency range in audacity is 0 Hz- 8000 Hz. In our case, sound frequency lower than
4000 Hz is a noise. Hence we will remove the noise.
Figure 3.4: Sound Spectrum after removing the noise
Audacity software applies Inverse Fourier Transform to achieve noise free sound.
View publication stats
4 .Conclusion
All waveform disturbed or scribbled observed in the universe, are just the sum of simple sinusoidal
of different wavelength. When a time-domain signal is represented in the frequency domain, i.e., as
a sum of sine waves, we can cure the noise by simply erasing all the highest frequencies and then
reconstituting the sound. In this paper, we discussed the theoretical aspect of Fourier Series used in
Discrete Signal Processing. We have demonstrated the theory of removing noise from a sound using
Audacity.
5 .References
[1] Sound \ Processing.org. (2019). Retrieved from https://processing.org/tutorials/sound/
[2] But what is the Fourier Transform? A visual introduction. (2019). Retrieved from
https://www.youtube.com/watch?v=spUNpyF58BY&t=700s
[3] Sound and Fourier series. (2019). Retrieved from
https://www.uio.no/studier/emner/matnat/math/nedlagte-emner/MATINF2360/v16/kompendiet/appl
inalgpythonchap1and2.pdf
[4] F. Alm, J., & S. Walker, J. (2019). SIAM (Society for Industrial and Applied Mathematics).
Retrieved from https://epubs.siam.org/doi/pdf/10.1137/S00361445003822
[5] Ahmed, N., & Natarajan, T. (1974). Discrete Cosine Transform. IEEE Transactions On
Computers.
[6] Oppenheim, A. (1975). Digital signal processing. Cambridge, Mass.: Massachusetts Institute of
Technology.
[7] Hollingsworth, M. (2019). Applications of the Fourier Series. Retrieved from
http://sces.phys.utk.edu/~moreo/mm08/Matt.pdf
[8] C. Courtney, A. (2019). Illuminating Fourier Series with Audacity. Retrieved from
https://arxiv.org/pdf/1211.4832.pdf
