Editorial board editor-in-chief

Acoustic investigation
Snoring is produced in the vocal tract, similarly to speech. Thanks 
to that analogy, existing techniques for speech analysis have 
been applied to evaluate snoring sounds. Snoring sound was 
precisely and easily detected with use of a condenser micro-
phone. Microphone was hung in front of the patient’s mouth at 
a distance of about 15-20cm. The signal was recorded and sent 
through the analogue-digital converter directly to the computer 
system and subsequent analysis was performed. 
Acoustic analysis of habitual and occasional snorers started 
with auditory analysis. For an individual who is an occasional 
snorer we can observe that his breath is very regular. In addition, 
only at the third hour of sleep regular sounds of snoring with 
very low intensity could be heard. On the other hand, a habitual 
snorer has an extremely diverse signal in the time domain which 
makes it dificult to interpret: throughout the entire night changes 
of snoring intensity could be heard and breathing is irregular. 
There are numerous techniques and methods employed to 
measure tsnoring. Acoustic analysis techniques give information 
on the mechanism, loudness, intensity and sites of obstruction of 
the upper airways. Snoring has been analyzed in the frequency 
and time domain and it was deined with the set of quantitative 
sound parameters (table 1).

---

T
elema
tics
38
Snoring as a sign of abnormality
Table 1. 
Techniques and methods employed to measure 
snoring [7]
Spikes in sound intensity of breathing 
Maximum snoring intensity
Mean snoring intensity 
Number of snorers ·h
-1 
of sleep 
Number of snorers·min
-1
of snoring
Power spectrum
Sonogram 
Formants structure 
LPC 
>45dB or >65dB
dB max
dB mean
SI
SF
The transformation of data from the time domain to the 
frequency domain was carried out by the Short-Time Fourier 
transform algorithm implemented in a PC software called “snore”, 
written in the Matlab programming environment. Sampling fre
-
quency of the analog-to-digital converter (44100 Hz) determines 
the maximum time duration of the sample. Frequency range of 12 
kHz can completely describe the snoring phenomenon.Snoring 
sounds were analyzed using the short-time Fourier transform 
(STFT) to determine the frequency and content of local sections 
of the samples. It can be described using the following equation:
STFT
T
x
= 
X(τ,f)
= 
∞
∫
−∞
x(t)w(t
−
τ)e
−2πft
dt
(1)
where 
w(t)
is the window function, commonly a Hamming win-
dow (width
N 
= 353 samples), centered around zero, and 
x(t) 
is the signal to be transformed. 
X(τ,f)
is essentially the Fourier 
Transform of 
x(t)w(t-τ)
, a complex function representing the 
phase and magnitude of the signal over time and frequency.
Time variation of the frequency spectrum is realized by divid-
ing the analyzed signal into short, overlapping segments (ig. 1). 
Signal in 10ms segments becomes stationary, so a short-time 
Fourier transform can be performed. After raising the resulting 
spectrum to the second power these segments can be combined. 
Time variation of the frequency spectrum is deined as square 
module of STFT [4, 5]. It can be described using the following 
equation:
G
x
(t, f ) 
= | STFT
x
= 
(t, f )
|
2
(2)
The STFT is a complete description of the signal and it is an 
important procedure for further analysis.
Waveforms of snoring events over a period of 10s (ig. 2) 
were analyzed. Depending on the snorer there correspond to 
from two to four respiratory cycles. The subjects have very regular 
respiratory cycles, unlike typical snoring patients. 
The frequency domain provides most important information 
from snoring sound, enabling power analysis and three-dimen-
sional graphs [1]. The averaged spectrum shape of snoring event 
is represented with values (Hz) of its formants. Different condi
-
tions in which subjects and patients snore can affect formants 
range. Examining snoring sound signal during sleep, energy was 
mainly concentrated in low frequencies, below 6000Hz. The main 
components lie in the low frequency range, at about 130Hz. The 
spectrum shows a fundamental frequency and formants structure. 
Also the frequency spectrum changes in every snoring event or 
during respiratory cycle (ig. 3 and 4).