STUDY ON HNR IN TRANSMITTED SOUND SIGNALS

Shouguo Zhao' Fanliang Bu2 Yihe Sun' Chengcai Han3

1 Institute of ~licroelectroiiicsiTsiiighua Liniversity, Bei.jiny, 100084, China

zhaos~r/~tsioehi~a edu.cn s i ~ n \ ; t i ~ ~ ~ t s i n e h t ~ a . e d u . c i i
li Sation I-abontoiy oii Machine Processing, Pe.hing Lniversity, Beijing I0087 I China
blJ@is. vkti.edd~m
3. h4cclioiiicol Engiiiccriiig Collcgc Xi'an Shiyou University, Xi'aii 51 0069 China

ABSTRACT laryngeal diseases.

Human voicc siplials arc complcx acoustic signals
In this article. the estimation of Haniioiuc to Noise
originated from aerodynamics, hydrodynamics:
Ratio (HNR) in transmitted sound signals with
biomedicine, neurology and acoustics in the vocal
wavelet transform \vas proposed. When subjects
tract and are not strictly periodic. The HNR in human
phonate sustained vowcls in breahy, falsetto, leakage
voice signals is a primar?; characteristic in diagnoses
and pressed mode in nornial loudness: the HNR in
for laryngeal diseases. When there are diseases in
human voice and transmitted sound signals was
vocal folds and larynx, the quantit?.. HNR in human
estimated and compared. The results indicated that
voice signals can image the degree of laryngeal
the NHR in transmitted sound signals could
diseases. HNR estimation can be accomplished in
accurately image vocal folds vibration characteristics.
two ways: (1) on a time domain in which
I t would be one of useful iiieaswements for clinical
HNR I s computed directly from the acoustic
laryngeal dsease diagnosis.
waveform; and (2) on a frequency-domain
Key words: transmitted sound: wavelet transfonii; in which HNR is computed from a transfomied
Hamioiiic to Noise Ratio representation of the waveform. A number of
modified techniques ['-13' h ave been proposed for
calculating HNR in time and frequency domain.
1. INTRODUCTION Because that there are period-to-period variations in
, fundamental frequency and amplitude modulation

Wlieu vocal folds vibrate i n vocal tract: human voice perturbatious in human voice signals, current HNR
signals are produced by the supra-glottal voice source estimation techniques in time and frequency domain
transmitting tluough supra-glottal vocal tract oral have many deficiencies. It made that results can't
ca\;it\. and lips. transmitted sound signals on the wall accurately image vocal folds vibration Characteristics
or thorax and on the neck are synchronously and laryngeal diseases.
produced by the sub-glottal voice source transmitting Because there are frequency modiilntion noise and
through sub-glottal vocal tract and lung-thorax amplitude modulation noise in human voice and
sections. We knew that human voice signals were transmitted sound signals, current methods in time
mainly determined by supra-glottal voice source, and frequency domain for HNR estimations produce
supra-glottal vocal tract and oral cavib. so the NHR more errors. In this article. the wavelet transform \vas
in voice sigials can't accurately image vocal folds used to estimate the harnionious and noise
vibration charactcristics and lavngcal discascs. coinponcnt['"l"'. HNR cstiniation with wavclct
Because of these deficiencies of the HNR in voice transform is free from amplitude and frequency
signals. HNR iu tmnsmittcd sound signals were put pcrturbation. The HNR in transmitted sound signals
fonvard for estimation vocal folds vibration is tested and investigated in different phonation
characteristics and for clinical lalyngeal diseases modes, including the normal, pressed. breathy:
diagiosip. Trausmitted sound sigials on V-slot below falsetto and some larygeal diseases. The results
ueck are not as oral voice signals through long vocal showed that the HNR in transmitted sound signals
tract, so the HNR in' transmitted sound signals is could accurately image the relations behveen vocal
better image vocal folds vibration characteristics and folds vibration characteristics and laryngeal diseases.

0-7~03-7902-n/03/si o
7.00 2003 IEEE. 5 80
 computer at a sampling rate of I O KHz and 12-bit
A/D resolution.
2. DATA ACQUISITION

Based on the producing principle of voice and

3. EXPERIMENT RESULTS
transmitted sound, we used data acquisition system as
Fig. 1. These subjects were divided into three groups. 3.1 Normal subjects HNR in a variety of vowels
The first group is that the health group consists of 20
In this article, the HNR in signals is given by
subjects (10 men and 10 women) with an age range
of 20-60 years old. Phonation tasks include: 0
I”En(/)’
N

phonating the sustained vowels [a:], [i:], [U:], [e:] and HNR = lOlog,,{xh(i)? 3-1
i=l i=l
lo:] at least three seconds in chest register with a
comfortable level, @phonating the sustained vowel Where h(i) is harmonious component and n(i) is
[a:] in the breathy, falsetto, leakage and pressed noise component.
modes with normal loudness. The second group
HNR estimations of voice and transmitted sound
consists of I O vocal folds hyperemia subjects (5 men
signals samples with wavelet transform methods
and 5 women) with an age range of 20-60 years old,
were accomplished. -Normal subjects HNR estimates
they were asked phonating the sustained vowel [a:],
in a variety of vowels had been given in Fig.2. It is
[i:]Ju:]: [e:] and [o:] respectively on the conditions of
found that voice sound HNR changes significantly,
nomial level. The third group includes 23 laryngeal
but transmitted sound HNR changes little in a variety
patients( 12 men and 11 women) with an age range of
vowels in chest register with a comfortable level.
20-65 years old. Among these there are 3 unilateral
When a normal subject phonates the sustained
paralysis subjects, marked PI: 3 partial laryngectomy
different vowels, the vocal fold vibration modes are
subjects, marked P2, 9 functional dysphonia subjects,
basically same. The super vocal tract, oral cavity and
marked P3> 5 nodule subjects, marked Pa, and 3
lips cause the great changes of NHR in voice sound
serious polyps of vocal folds, marked P5. They were

-
signals, so using HNR of transmitted sound signals
asked phonating the sustained vowel [i:] respectively
can decrease effects of super vocal tract, oral cavity
on the conditions of normal level.
and lips. It could image the vocal folds vibration
characteristics.

dB 40
35 c
20
15 -- Oral sound
Trmrmitfcd
saund

Fig. 1. Data acquisition system

Voice and transmitted sound signals samples were
l5
o
0 1
I ,Transmitted
sound

[a:] Ci:l [U:] [e:] CO:]

simultaneously recorded using condenser microphone
All recordings \*ere low-pass filtered (fC=5KHz), Fig.:! Normal subject HNR in a variety of vowels in
high-pass filtered (fh=70Hz) and digitized into a chest register

581
3.2 Normal subject HNR in a variety of phonation vocal folds hyperemia subjects phonate sustained in a
modes variety of vowels, HNR in transmitted sound signals
have less changed than HNR in voice sound signals.
When normal subjects phonate the sustained vowel
The HNR in transmitted sound signals can distinctly
[a:] in breathy, falsetto, leakage and pressed modes
image the vocal vibration characteristics changes by
with normal loudness, HNR results in voice and
vocal folds hyperemia. The dynamic range of the
transmitted sound signals are shown in Tablel. It is
HNR in transmitted sound signals is less than the
found that the HNR in voice and transmitted sound
dynamic range in voice sound. In this way, The HNR
signals are all changed with the phonation modes.
in transmitted sound signals can more accurately
The NHR in transmitted sound is higher than the
image the vocal folds vibration characteristics.
HNR in voice sound signals. The NHR in transmitted
sound signals can image the changes of vocal folds
vibration modes, which is consistent with the NHR dB 30
changing tendency in voice sound signals. In falsetto 25
mode, the HNR in'transmitted sound signals is the
highest, then in order of breathy, pressed and leakage,
perhaps, because of higher fundamental frequency
and more perfect periodic vibrations. The results are
coincident with the pathology and physiology of
different vocal folds vibration modes.

TlWlSrmtted
Phonation Voice sound
Fa ( H z ) sound HNR
niodes HNR (dB)
(dB) Fig.3 Vocal folds hyperemia subjects HNR in a variety
of vowels
Pressed 70 26.14 28.04
3.4. Laryngeal disease subjects HNR
Breathy 120 30.54 36.29
When subjects have laryngeal disease, vocal folds
Leakage I05 23 26 24.48
vibrations are irregular and glottis closed is not
Falsetto 300 35.49 42.62 perfect. In this way, there are stronger fundamental
frequency and amplitude modulation perturbations.
Conventional methods in time and frequency domain
don't give accurate estimation for HNR in voice and
3.3. Vocal folds hyperemia subjects HNR in a transmitted sound signals, so wavelet transform
variety of vowels methods can accurately estimate the NHR in voice
and transmitted sound signals. Experimental results
These vocal folds hyperemia subjects HNR are
are shown in Table 2. On the condition of sustained
shown in Fig.3. In vocal folds hyperemia subjects,
vowels, these HNR in voice and transmitted sound
the HNR in voice and transmitted sound signals both
signals of P,, P2, P3, P4 and Psgroup are lower than
are reduced in some extent. From these results. when
the normal subjects.

582
 These HNR cstimates in transmitted. sound signals diagnose. It is potentially easier to apply in voice
are also higher than the HNR estimates in voice research and clinical evaluations of voice disorders
sound. HNR of laryngeal disease subjects can be and laryngeal diseases.
accurately estimated with wavelet transform method.
Experimental results further verified the validity of 5. ACKNOWLEDGEMENTS
HNR estimations in transmitted sound and are
coincident with its physical and physiological The research work in this paper was supported by
meanings. The HNR in transmitted sound signals can National Natural Science Foundation of China
image the vocal folds vibration characteristics of (No.60275018), (No.30070212) and (No.60236020)
different laryngeal diseases.
References
Group Nuniber Voice sound Transmitted sound
HAIR (dB) HNR (dB) [l] .E.Yumoto, W Gould, and T.Bear.
I‘
Harmonic-to-noise ratio as index of the degree
p, 1 3 13.16 15.14
of hoarseness” , J. Acoust. Soc. Am., Vol. 71.
10.31 No.3. PP. 1544-1550, 1982
[2]. Y.Y.Qi. “Time normalization in voice analysis”.
13.78 16.48
J. Acoustics. Soc. Am., Vol. 92. No.4 PP.
PA 16.49 20.19 2569-2516, 1992
1 Pi I 3 I 11.32 1 14.50 I [3]. Y.Y.Qi. et al. “Minimizing the effect of period
determination on the computation of amplitude
Table-2 Laryngeal disease subject HNR . perturbation in voice” . J. Acoust. Soc. Am., Vol.
97. No.4 PP. 2525-2532, 1995
4. CONCLUSIONS [4]. H.Kasuya, et al. “Normalized noise energy as an
acoustic measure to evaluate pathologic voice. ”
J. Acoust. Soc. Am.,Vol. 80. No.2 PP. 1329-1332,
We have presented a new approach to quantify the 1986
amount of vocal folds vibration characteristics by [ 5 ] . H.Muta, T.Bear. “A pith-synchronous analysis of
HNR in transmitted sound signals with wavelet hoarseness in running speech.” J. Acoust. Soc.
transform that is robust against irregularities of the Am., Vol. 84. No.2 PP. 1292-1301, 1988
glottal oscillations. For comparison, the reformed [6]. Cox et al. “ Technical considerations in
time-domain method for HNR estimates has been computation of spectral harmonics-to-noise
tested. The HNR estimations in transmitted sound ratios for sustained vowels”. J. Speech Hear. Res.
signals with wavelet transform is a reliable measure Vol. 32. No.2 PP. 203-218. 1989
for the relative noise level and effectively image the [7].De Korm. “ A cepstrum-based techniques for
vocal. folds vibration characteristics even in the determining a harmonics-to-noise ratio in speech
presence of strong aniplitude and periodicity signals” . J. Speech Hear. Res. Vol. 36. No. 2.
perturbation We gave HNR results in transmitted PP.254-266. 1993
sound signals in different vowels, pressed, falsetto, [S].Y.Y.Qi, et al. “ Temporal a n d spectrum
breathy phonation modes and typical laryngeal estimation of harmonic-to-noise ratio in human
diseases. Through a lot of investigations, the HNR in voice signals” . J. Acoust. Soc. Am., Vol. 102.
transmitted sound signals would be one of the useful NO. 1. PP.531-543. 1997
measurements for clinical laryngeal diseases

583
 [9].D. Michaelis, et al. “Glottal-to-noise excitation
ratio--a new measure for describing pathological
voices” . ACTA. Acoustics, Vol. 83. No. 3.
PP.700-706. 1997
[lO].P.J.Mulphy. “Perturbation-free measurement of
the harmonics-to-noise ratio in voice signals
using pitch synchronous harmonic analysis.” J.
Acoust. Soc. Am. Vol. 105. No. 5.
PP.2867-2881. 1999
[I I].G.de Kroni. “ A cepstrum-based technique for
determining a harmonic-to-noise ratio in speech
signal.” J. Speech Hear. Res. Vol. 36. No.2 PP.
203-218, 1993
[ 12].H.Kasuya, Yendo, S.Saliu. “ Novel acoustic
measurements of jitter and shimmer
characteristics from pathological voice. ”
UROSPEECH ‘93 Berlin, PP.1973-1976. 1993
[13].Wang supin Wan mingxi et al I

“ Harmonics-to-noise ratio analysis of

pathological speech signals” Joumal of XI’AN
Jiaotong University Vo1.30. No.9. PP.53-58. 1996
[14].S.Mallat, S.Zhang. “Characterization of signals
from multiscale edges.” IEEE Transactions on
Pattem Analysis and Machine Intelligence, Vol.
38, NO. 2: PP.617-643, 1992
[ 151.S.Mallat,W.L.Hwang, “Singularity detection and
processing with wavelet. ” IEEE Transactions on
Information Theory>Vol. 38, No. 2, PP.617-643,
1992

5 84