5. Sound Processing

Question 1

What are the basic parameters of sound?

Answer 1

The sound that we hear around us is the type of energy made by the vibration that travels through the air or any other medium and can be heard when it reaches a person’s ear.

The characteristics of sound are as follows:
- Pitch: A note has a higher pitch when the frequency is high and a note of low frequency has a low pitch.

• Loudness: a sensation of how strong a sound wave is at a place. It is always a relative term and is a dimensionless quantity. Loudness is measured in decibel (dB) and depends on the amplitude of the vibration.
• Quality: The word timbre also describes the term quality and describes the characteristics of sound which allows the ear to distinguish sounds which have the same pitch and loudness.

Question 2

What does Fourier analysis mean?

Answer 2

The Fourier analysis produces a decomposition of any signal into a series of sinusoidal waves with different frequencies, amplitudes, and phase shifts

Question 3

What is FFT? What are the inputs and outputs?

Answer 3

Fourier analysis of digital signals is done in practice by using the discrete Fourier transform.

The discrete Fourier transform (DFT) is an algorithm that transforms a time-domain representation of a digital signal into a frequency-domain one

The fast Fourier transform (FFT) is a widely used, fast, and efficient algorithm for the DFT

Black Box:

• Two inputs:
  
  1. Vector containing digitized signal samples
  2. Number of samples
• Output:
  
  • Vector that contains amplitudes

Question 4

How are signals represented in frequency domain? (sinusoidal signals, sum of sinusoids, periodic signals, short time Fourier analysis, spectrogram)

Question 5

What does windowing mean? Why do we need it?

Answer 5

Segmentation or windowing is the process of splitting a signal into time chunks
- Signal is cut into short time chunks called windows or frames
- FFT can be applied to signals that are considered reasonably stationary and almost periodic
-
Problem: by cutting a signal into time frames, certain features might be lost because part of the signal might belong to one segment and the other part to the next one

Solution: overlap successive frames

Question 6

How can we filter a sound signal? (time/frequency domain)

Answer 6

The moving (or running) average filter is used to eliminate random noise by taking the average of a few consecutive measurements in time domain. Thus, it calculates each point in the output signal by averaging consecutive input samples. This is used when the noise course is unknown.

A frequency-selective filter only allows selected frequencies while blocking the rest. This is used when the source of noise is known, such as in an ECG, and thus always produced a superior output.