790 Proceedings of the Royal Society of Medicine 42

the interpretation of that particular form of response and stimulation. In the tests which are being
'Carried out to get an idea of the most desirable response for hearing aids this apparent ability of the
auditory system to adapt itself to a particular response sets us various problems. One cannot be sure
that articulation from what might otherwise be the optimum response is not influenced by the listener
'being unaccustomed to it. Perhaps the most crucial tests will be those with patients who have not
heard for a long time. At the present time the most pressing requirement is that of power efficiency.
Should this become solved the question of frequency response can then be further investigated.
ACKNOWLEDGMENT
The work discussed in this paper was carried out as part of a programme of the Wernher Research
Unit on Deafness maintained by funds generously provided by The Alexander Pigott Wernher
Memorial Trust to whom grateful acknowledgment is made.
REFERENCES
AYERS, E. W., and MORTON, J. Y. (1951) Acustica, 1, 109.
BARANY, E. (1938) Acta Otolaryng. (Supp. 26).
BEIKESY, G. v. (1932) Ann. Phys. Lpz., 13, 111.
GuiLD, S. R. (1936) Ann. Otol., etc., St. Louis, 45, 736.
Spec. Rep. Ser. med. Res. Coun., Lond. (1947) No. 261. Hearing Aids and Audiometers, p. 68.
TUMARKIN, A. (1946) J. Laryng. Otol., 61, 473.
WEVER, E. G., and LAWRENCE, M. (1952) Proc. Nat. Acad. Sci., 38, 133.

Masking in Pure-tone Audiometry

By P. DENES, M.Sc., A.M.I.E.E.
Phonetics Dept., University College, London
and
R. F. NAUNTON,1 M.B., B.S.
Wernher Researchi Unit on Deafness, Medical Research Council, London
IT is a common experience that when any sound is applied to an ear it decreases the ability of that
ear to perceive other sounds. If, while one sound is being applied to an ear, a second sound is gradually
increased in intensity until the first sound can no longer be heard, the first sound is said to be masked
by the second sound (Fletcher, 1929).
This commonplace observation can be put to good use in clinical audiometry where it is often
imperative while testing one ear to prevent perception of the audiometer tones by the other ear.
In air-conduction audiometry, for example, if there is a large difference between the thresholds of
the two ears, the sound applied to the deafer ear may be perceived by the opposite better ear, and the
familiar shadow audiogram will result. Only by masking the hearing of the intruding better ear can
this effect be prevented. Similarly in bone-conduction testing, even in a case where one labyrinth has
been deliberately destroyed by surgical means, the sound stimulus will be perceived whichever mastoid
the bone conductor is placed on, because even when the bone conductor is applied to the deafened
side the sound will be transmitted to the good ear, almost without loss. In bone-conduction audiometry
the untested ear should always be masked. Two other types of hearing test where, in many cases,
accurate results can be obtained only when the untested ear is masked are speech audiometry and
recruitment detection. . r
The importance of masking as a necessary adjunct to monaural hearing testing has long since
been recognized and a variety of ingenious methods have been tried. One of the earlier and best-
known examples is the noise box of Barany, a device still in everyday use. Other suggestions include
the use of streams of air or water directed into the ear to be masked. Others have proposed for use
as a masking sound a pure audiometer tone of the same frequency as the audiometer test tone; and
at least one commercial audiometer has incorporated this method of masking. Many modern types
of audiometer use mains hum for masking, while quite recently Zwislocki (1951) has utilized as a
masking sound the audiometer tone itself, modified in a fashion to be described later, which is intended
to make it distinguishable from the audiometer test tone.
Before discussing the relative merits of these methods and of our own suggestions, let us examine
the features theoretically desirable in an ideal masking sound. Two different masking sounds which
we may designate A and B may be equally effective in masking a given audiometer tone, but none the
less the masking sounds A and B may differ markedly in loudness. It is obviously undesirable to have
an unnecessarily loud masking sound, and we may therefore at this stage introduce a new term, the
Masking Efficiency of a sound. This we define as the relation between a sound's ability to mask and
'Holder of a grant for research awarded by the Medical Research Council.
43 Section of Otology with Section of Laryngology 791
its loudness. A sound with high masking efficiency is one with good masking ability but minimal loud-
ness. This then is the first desirable feature of an ideal masking sound, it must be efficient. A second
requirement is that the masking sound must, for two reasons, be readisy distinguishable from the
audiometer test tone. First, it is a great advantage to be able to describe the two sounds to the test
subject in terms which will not confuse him, and secondly, the test subject must from the description
given him be able to recognize and distinguish the two sounds when he hears them simultaneously.
This argument is all the more important when testing deaf subjects suffering from diplacusis.
The third, but by no means least important, desirable feature is that any apparatus required shall
be as inexpensive as possible, and simple to operate.
Recapitulating, the three desirable features in the light of which we shall examine the usefulness of
the various masking sounds are: (1) Masking efficiency; (2) distinguishability of masking sound from
audiometer tone; (3) simplicity in use and manufacture.
It is scarcely necessary to enumerate the many objections to the use in audiometry of such masking
devices as the B1rany noise box and jets of air or water. They are unpredictable in effect and awkward
in use.
The objection to the use of a pure-tone masking sound having the same frequency as the audiometer
tone -is that there will be grave difficulty in distinguishing the two sounds. This difficulty can be
decreased by introducing a difference between the frequencies of the masking and the masked tones.
But as the difference between the two frequencies is increased so the efficiency of the masking sound
decreases, and if the frequency difference is made large enough to permit ready distinction, the efficiency
will be very low, or in other words the masking tone will have to be very loud to produce a given
masking effect.
The masking of one audiometer tone by means of another tone of markedly different frequency is
the principle underlying the use of mains hum as a masking sound. Here the masking sound, provided
by the mains electricity supply, is a pure or near-pure 50 c/s tone. The apparatus required for the
generation of mains hum is extremely simple, and there is no difficulty in distinguishing the masking
sound from the audiometer test tone; but the hum will frequently need to be unpleasantly loud in
order to produce a relatively small masking effect. This principle is used, however, with some slight
modifications, by several audiometer manufacturers, one of whom claims that the masking effect of
this type of noise is uniform for all the audiometer test-tone frequencies, and implies that the effect
is independent of the type of deafness present in the ear to be masked. But our investigations of this
type of noise have demonstrated its extreme unreliability. With a given amount of masking noise the
masking effect produced upon a group of deaf subjects at each of the audiometer test-tone frequencies
was found to vary almost at random by as much as 40 db., even after allowing for their- different
degrees of deafness (Fig. I).
30

25

20

-~~~~l5 ~~~~~~60
o cls
5000 and 4000 c,
2000 c,/
E
1n0 ~40 soo CLS .

z
5 Lu 20 250 c/rs.
0 .
~0 __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

-5 0 5 10 15 20 25 30 35 E 0 j1 20 30 40 50 60
Masking Effect in decibels. Sensation Level of White Noise in decibels.
FIG. 1.-The masking effect (decibels FIG. 2.-The masking effect (decibels rise in test-tone threshold)
rise in audiometer test-tone threshold) produced by white noise upon a series of normal listeners, at
produced by modified mains hum upon various test-tone frequencies and at various sensation levels
a series of deaf listeners, at various test- (decibels above the tested subject's threshold for white noise).
tone frequencies and at a sensation level
of 30 decibels (i.e. 30 decibels above the
deaf subject's threshold for the mains
hum).
792 Proceedings of the Royal Society of Medicine 44
The presence in this noise of numerous harmonic components makes it impossible accurately to
predict its masking effect upon a deaf subject whose pure tone threshold audiogram is not flat.
Another type of masking noise we mentioned is that proposed by Zwislocki. In this method part
of the audiometer output is led to the tested subject in the usual fashion, but the remainder of the
output is modified with the help of mains hum in such a way that what was a single pure tone becomes
a collection of different tones ofslightly differing frequencies. This relatively simple manceuvre produces
a masking sound which is efficient (in other words to produce a given masking effect it does not have
to be excessively loud) but our experience has shown that many tested subjects, deaf and normally
hearing alike, experience considerable difficulty in distinguishing the masking sound from the un-
modified audiometer tone. We have already emphasized that for ease of testing it is desirable that the
masking sound be readily distinguishable from the audiometer tone.
At this point it has become apparent that there may be room for improvement upon the masking
methods we have reviewed. The major weak points in the masking sounds we have considered have
been either poor masking efficiency, or difficulty of distinction, or a combination of both these faults.
The simplest way of providing ready distinguishability is to ensure that the masking sound shall
always appear to be a definite noise when compared with the audiometer tone. Now a readily available
type of noise having no resemblance whatsoever to a pure tone is white noise or random noise. It
consists of a mixture of tones in which all audible frequencies are present in equal amounts, and its
production involves only the most elementary electrical apparatus. Used without any modification
white noise does exert a masking effect and it is readily distinguishable from all the audiometer tones
(Fig. 2).
But the masking effect of white noise depends to an unpredictable extent upon the shape of the
tested subject's audiogram, and it is at best very inefficient. These two objections to the use of white
noise as a masking sound can be overcome by filtering it in such a way that only those frequencies
which are near to the frequency of the audiometer test tone remain. Thus, instead of containing all
audible frequencies it will contain a band of relatively few frequencies centred around the test-tone
frequency (Fig. 3), and by using a variety of different filters the frequency at which the band centres
can be varied. Such a filtering process will result in a masking sound much more efficient than un-
modified white noise and much less dependent in its effect upon the shape of the audiogram. But if
this process of filtering is carried too far and the band of frequencies left after filtering is too narrow,
the resulting filtered noise will assume a tonal quality.
A number of tests were made to find out how far this filtering process can be carried; that is to say,
how wide must the noise band remain in order to ensure that the noise does not lose its essential
noise-like character and become a tone. We found that the white noise could be restricted to a band
one-third of an octave wide and could still, in practice, be recognized as a noise by the vast majority
of listeners. Any narrower band produced a sound with a readily recognizable tonal rather than noise
quality. These observations led us to investigate further the suitability of one-third of an octave wide
noise bands for use as masking sounds. Five of them centred about 250, 500, 1000, 2000 and 4000 c/s
respectively were tested upon a group of normal and deaf subjects. It was found that the masking
effect of all five noises upon their respective test tones was consistent and could be predicted in practice
with accuracy; irrespective of the shape of the tested subject's audiogram (Fig.4).
Further, determinations of the loudnesses of these noise bands, and of the other types of masking
sound that we have mentioned, by means of the familiar loudness balance technique showed that,
for a given masking effect, the one-third of an octave wide noise bands are as efficient as the best of
the other masking sounds (Fig. 5).
TABLE I.-SHOWING LOUDNESSES IN SONES OF VARIOUS MASKING SOUNDS WHEN A 20 DECIBEL
MASKING EFFECT IS PRODUCED
(Relative Masking Efficiency =L
Loudness iin ssones
Frequency of audiometer test tone to be masked
Type of masking sound.. 250 c/s 500 c/s 1000 c/s 2000 c/s 4000 c/s
Modified mains hum .. .. (beats) 10 8 19 57
Zwislocki noises .. .. .. 27 4 3 0.6 06
White noise .. .. .. 200 40 15 27 15
Third of an octave noise bands.. 2 1.0 0-7 1.0 10

The first of the three desirable features which an ideal masking sound should display is efficiency,
i.e. the sound should provide good masking effect but be as quiet as possible. On this score, as we
have just pointed out, the one-third of an octave noise bands are the most suitable of the various
Jounds we have reviewed. So far as the second criterion is concerned, these third of an octave bands
are, as our experiments have indicated, noises rather than tones, and will therefore be readily dis-
tinguishable from the audiometer tones which they are intended to mask. There remains for consider-
ation the third criterion, namely that the masking method shall be simple in use and require a minimum
45 Section of Otology with Section of Laryngology 793

_a

100 .o
I-,
c)
60 I.
..
.........
.......................... ....... .... .......
......
I -v *40
a.', , ..........
Intensity....
,n te slty ,., :. s. :. '., .: , . .S:. fi:. . ,.s USasaSgSs. .
Scale. ( l.)--... :::::.:: :-.::::.:::
::.:::::B..
...........
(a) *
" 20
.'...'.....'" .............' ::'"
::

100 Audio-frequency range. (F *) -I' .

'L _

0 10 20 30 40 50 60
Sensation Level of Masking Sound in decibels.
FIG. 4.
(b)
01
2f (F.) FIG. 3.-Schematic representation ofthe constitutiolf
100 of unmodified white noise (a), and of the effect upon
its constitution of filtering it so that (b) an octave
wide noise band, (c) a third of an octave wide noise
band, and (d) a sixth of an octave wide noise band
remain.

(I) (c)
FIG. 4.-Masking effect (decibels rise in threshold
f 2f (F.) of audiometer test tone) produced by various masking
100 sounds upon a series of normal listeners, at various
test-tone frequencies, and at the various sensation
levels shown (decibels above the tested subjects'
threshold for the masking sound). Masking sounds
thus treated: (a) Zwislocki noises at 2000 and 4000
c/s, and third of an octave noise bands centred at 250,
(d) 500, 1000, 2000 and 4000 c/s. (b) Zwislocki noise
(I) u at 500 c/s. (c) Modified mains hum on 1000 c/s.
(d) Zwislocki noise at 1000 c/s, and modified mains
f 2f (F.) hum on 500 c/s. (e) Zwislocki noise at 250 c/s.
(f) modified mains hum on 2000 c/s. (g) Modified
FIG. 3. mains hum on 4000 c/s.

70 FIG. 5.-Loudness balance graph

of results obtained with normal
0)
v
listeners, relating the sensation level
4- (decibels above the tested subject's
threshold) of a 1000 c/s audiometer
C
test tone, to the sensation level to
0 which various masking sounds
0
0 50 must be raised in order that they
-0 may be of the same loudness as
the 1000 c/s test tone (i.e. loudness
40 of various masking sounds in
-_ Phons at various sensation levels).
r_,u Masking sounds thus treated: (a)
.C 0
Modified mains hum, andZwislocki
w
.- .E 330 noise at 250 c/s. (b) Third of an
C
0) octave noise band centred at 250
c/s. (c) White noise. (d) Zwislocki
noise at 500 c/s. (e) Third of an
15 20 25 30 35 40 octave noise band centred at 500,
2000 and 4000 c/s. (f) Zwislocki
Sensation Level of Masking Sound in decibels, noise at 1000 c/s. (g) Third of an
producing equal sensation of loudness. octave noise band centred at 1000
c/s. (h) Zwislocki noise at 4000 c/s.
FIG. 5. (j) Zwislocki noise at 2000 c/s.
794- Proceedings of the Royal Society of Medicine 46'
of extra apparatus. So far as simplicity in operation is concerned, it is merely necessary to know the
hearing loss of the ear to be masked for the tone which is to be obscured. Supposing this hearing loss
to be 20 db. and the masking effect required 30 db., then the masking sound attenuator must be set
to the 20 + 30, that is, the 50 db. position. It is scarcely possible to devise any simpler method than
this. The extra apparatus required is a gramophone reproducer which will replay recordings of the
necessary third of an octave noises, plus an attenuator for the control of the intensity of the masking
stimulus. It is felt that many hearing testing departments will already have a speech audiometer which
will admirably serve,4he same purpose; in such a case the only additional cost will be that of the
gramophone recording of the noises which we hope will be available in the near future. In the case
of many pure-tone audiometers it will be possible for the manufacturers to incorporate this type of
masking sound, the only additional expense being that of the record player.
As an alternative to the gramophone reproducer, it is possible to include a noise generator and the
necessary filters in the pure-tone audiometer. In this event the audiometer control which selects the
frequency of the test tone will, at the same time, select the correct masking noise band. The question
of providing this added circuitory at low cost and without adding materially to the bulk of the pure-
tone audiometer is at present being investigated. Information on these points, together with details
of the electrical characteristics of the filters and other equipment used in these experiments, will be
published in full elsewhere.
Summarizing the points we have attempted to make, these third of an octave bands provide masking
sounds of high efficiency which are easily distinguishable from the test tone they are intended to mask.
They are simple in operation and independent of the tested subject's type of audiogram, and last but
not least, the necessary apparatus is simple to manufacture and can be incorporated at low cost in
mxsost commercial audiometers.
REFERENCES
FLETCHER, H. (1929) Speech and Hearing, London.
ZwIsLocKI, J. (1951) Acta oto-laryng. Stockh., 39, 338.
Mr. A. Tumarkin asked whether it might not be more efficacious to use third of an octave noise bands
centred at a frequency slightly below rather than at the test tone frequency.
1In reply, Mr. Denes and Mr. Naunton agreed that such a measure would probably produce a masking
sound of slightly greater efficiency, but as all the available noise filters centre at the customary round
figure audiometer test-tone frequencies, and as any variation from this common practice would be more
expensive, it was felt that the slightly less efficient compromise was justified.

The Comparative Anatomy of the Labyrinth

An Example of the Evolution of a Special Sense Organ
By OLIVER GRAY, M.D., Haslemere
OPPORTUNITY was given, at this meeting, of seeing a number of specimens illustrating the evolutionary stages
in the development of the labyrinth or internal ear, commencing with the simple organ found in the Cyclo-
stomes (Hag-fish and Lamprey), to the elaborate structures which exist in Fishes, Amphibians, Reptiles,
Birds and Mammals. This makes a most fascinating study, for the story is a fairly complete one, and,
i,qi*ntally, deals with the most beautiful entity in vertebrate anatomy. A glance at the photographs of the
kby`intfis of Man (Figs. 1 and 2) and of the Pig (Fig. 3) will abundantly support the latter contention. Such
$*imens require a certain amount of patience and skill to prepare, and are only obtained as a result of a
Proess which takes at least three months.
.-4ifficvlty which often confronts the uninitiated lies in the significance of the membranous or otic labyrinth,
"
4istinct from the osseous or periotic labyrinth which surrounds it and contains perilymph. The capsule,
mQsay bodne, which enclosed the membranous labyrinth of the fish, invests the organ much more intimately
phibians, reptiles, birds and mammals; so that the semicircular ducts come to lie within the semi-
-am
cficul4rc,Znals, the utricle and saccule within the vestibule, while the endolymphatic tube of the cochlea, called
the. *4aus cochlearis or scala media, becomes surrounded, on either side, by the perilymph spaces known as
the scala tympani and scala vestibuli.
'TPerhaps another helpful way of explaining how the membranous labyrinth comes to lie, more or less, sus-
pende4d w'ithin the osseous labyrinth can be offered. In the human embryo the membranous labyrinth,
,ctoderm'ic in[ origin, has almost completed its final shape by the 40 mm. stage. It is only now that rifts begin
to appear in the surrounding mesenchymatous tissue. These rifts eventually coalesce to form the perilymphatic
spaces which constitute the osseous or periotic labyrinth, the completion of which occurs about the middle of
fcetal life.
-t can be argued that the labyrinths of the hag-fish and lamprey do not represent an early stage in evolutionary
development, but are degenerate forms of a more elaborate mechanism, which may have existed in earlier
times. The cyclostomes are certainly very degenerate creatures in many respects, and it is curious that, being
of the same family, their labyrinths are so dissimilar. This may, however, be accounted for by the fact that
each has a distinct and ancient origin; each, as it were, with a long, independent twig coming from the
Marsipobranch stem, very early in fish evolution. Again, the related fossil Cephalaspids, which are not
supposed to be degenwrate, had two semicircular ducts, like the lampreys (Stensio). Things must have their