null null

Development of the TPB
Development of the Test de Phrases dans le Bruit (TPB)
Élaboration du Test de phrases dans le bruit (TPB)
Josée Lagacé
Benoît Jutras
Christian Giguère
Jean-Pierre Gagné
Abstract
Josée Lagacé, PhD
École d’orthophonie et
d’audiologie,
Université de Montréal,
Centre de recherche du
CHU Sainte-Justine
Montreal, Quebec
Canada
Benoît Jutras, PhD
École d’orthophonie et
d’audiologie,
Université de Montréal,
Centre de recherche du
CHU Sainte-Justine
Montreal, Quebec
Canada
Christian Giguère, PhD
Programme d’audiologie et
d’orthophonie,
Université d’Ottawa
Ottawa, Ontario
Canada
Jean-Pierre Gagné, PhD
École d’orthophonie et
d’audiologie,
Université de Montréal
Centre de recherche de
l’Institut gériatrique
universitaire de Montréal
Montréal, Quebec
Canada
The Test de Phrases dans le Bruit, which consists of five French lists of 40 recorded sentences
and a speech babble, was developed for use in evaluating speech perception in noise. The
development of the sentence material was based on an approach that had previously been employed for the Speech Perception In Noise test. The key word familiarity of the sentences was
tested, as well as the intelligibility in noise. Measures were also performed to obtain equivalent
difference of scores between the high and low predictability sentences across the lists. Based
on the results obtained with a subset of adult participants, it is believed that the sentence list
sets that evolved from this work have the characteristics to be useful for the exploration of the
underlying auditory and/or language-based origins of speech perception problems in noise
among the Canadian French population. However, the present findings should be interpreted
with caution as only individuals with normal hearing function participated in the experiments.
The results may not apply to individuals with speech perception problems in noise. Additional
evaluations of the psychometric properties of the test must be performed before its clinical
application. Nevertheless, these preliminary findings suggest that further development of the
Test de Phrases dans le Bruit is warranted.
Abrégé
Le Test de phrases dans le bruit, qui est composé de cinq listes de quarante phrases enregistrées
en français et d’un bruit de verbiage, a été conçu pour évaluer la perception de la parole dans
le bruit. L’élaboration des phrases a été effectuée en suivant une approche similaire à celle du
Speech Perception in Noise (SPIN). La familiarité du mot clé de chaque phrase a été vérifiée, ainsi
que le degré d’intelligibilité dans le bruit. Le niveau de prévisibilité des phrases a été mesuré
afin de s’assurer que la différence de performance entre les phrases hautement prévisibles et
faiblement prévisibles soit équivalente entre les listes. D’après les résultats obtenus avec un
sous-groupe de participants adultes, on croit que les listes de phrases mises au point avec cet
essai pourront être utiles à la recherche sur l’origine des problèmes auditifs ou linguistiques
sous-jacents de perception de la parole dans le bruit parmi la population canadienne-française.
Cependant, les résultats actuels devraient être interprétés avec prudence, car seulement des
personnes avec une acuité auditive normale ont participé aux expériences. Les résultats pourraient ne pas s’appliquer aux personnes souffrant de problèmes de perception de la parole dans
le bruit. Des évaluations supplémentaires des propriétés psychométriques du test doivent être
effectuées avant son application clinique. Néanmoins, ces résultats préliminaires suggèrent que
la poursuite de l’élaboration du Test de phrases dans le bruit est justifiée.
Key words: speech in noise tests, language-based competencies, linguistic context
261 Canadian Journal of Speech-Language Pathology and Audiology - Vol. 34, No. 4, winter 2010
Development of the TPB
M
a ny i n d iv i d u a l s re p o r t d i f f i c u l t y
understanding speech in noise. For some
of them, their speech perception problems in
noise can be explained by their audiogram. For others, the
underlying nature of their difficulties is not as obvious. In
these cases, a better understanding of the listening problems
may improve service delivery.
The Speech Perception In Noise (SPIN) test was
originally developed to assess how well individuals with
acquired peripheral hearing loss utilize contextual linguistic
information to understand speech in noise (Elliott,
1995; Kalikow, Stevens & Elliott, 1977). The original test
materials consist of ten tape-recorded lists of 50 sentences
mixed with a twelve-talker speech babble. When the
SPIN test is administered, the listener is asked to repeat
the final word (key word) for each sentence. In each list,
half of the sentences are highly predictable (HP) as they
contain contextual linguistic information that facilitates
the identification of the key word (e.g., The candle flame
melted the wax). The other half of the list is composed of
low predictability (LP) sentences (e.g., Paul can’t discuss the
wax), which contain little contextual linguistic information
(Kalikow et al., 1977).
The SPIN test was developed on the premise that speech
perception involves at least two types of processes: 1) the
auditory processing of the signal and, 2) the languagebased processing of that information (Kalikow et al., 1977).
According to Kalikow et al. (1977), the recognition of the
final word of the HP sentences can be accomplished through
one or both of these operations, while the recognition
of the LP sentences key word depends mainly on the
auditory processing of the signal. The level of the babble
noise at which the test is conducted can be varied while
presenting the different lists of the SPIN sentences. This
test manipulation is relevant for determining the extent
to which responses for each type of sentences are affected
by the signal-to-noise ratio (SNR; Kalikow et al., 1977).
Since the two types of sentences of the SPIN test only
differ by the semantic and syntactic content, it is possible
to determine the extent to which the listener benefits from
the contextual linguistic information by analyzing the
difference of the performance for the HP and LP sentences
(Kalikow et al., 1977). Although the use of linguistic
contextual cues is only one component of the top-down
processing involved in the speech recognition process, it
is at least possible to measure this listener’s competency
with the SPIN test, which is not the case with the other
available speech in noise tests.
The original version of the SPIN test and the SPIN-R
test (the revised version of the SPIN test by Bilger, Nuetzel,
Rabinowitz & Rzeczkowski, 1984) have been used in many
studies to explore the underlying origins of the speech
perception problems in noise. For instance, it has been
employed in studies conducted among populations of
younger and older adults with normal hearing sensitivity
thresholds (Dubno, Ahlstrom & Horwitz, 2000; Humes,
Burk, Coughlin, Busey, & Strauser, 2007; Kalikow et al.,
1977; Pichora-Fuller, 2008; Pichora-Fuller, Schneider
& Daneman, 1995), adults with permanent hearing
impairment (Bilger et al., 1984; Schum & Matthews, 1992),
as well as adults with learning difficulties (Elliott & Busse,
1987). According to the results obtained with the SPIN test,
the speech understanding difficulties experienced by these
populations were related to underlying auditory deficits.
On the other hand, comparisons of the results obtained by
native listeners (listeners who learned American English
from birth) and non-native listeners (listeners who learned
American English later in life) on the SPIN test have lead
to different outcomes. The results revealed that the levels
of noise at which speech is intelligible are significantly
higher for the native listeners compared to the non-native
listeners (Bradlow & Alexander, 2007; Florentine, 1985;
Mayo, Florentine & Buus, 1997). It was also observed that
the benefit from linguistic context is significantly greater for
the native listeners compared to the non-native listeners.
The SPIN test provides a method of delineating the
relative contribution of the auditory and the language-based
function involved in speech understanding in noise. At this
point, there is no test available in Canadian French that is
comparable to the SPIN test. A simple translation of the
SPIN test sentences would not have been valid because of
the differences in the linguistic structure and vocabulary
between English and French. It was therefore necessary to
develop a French adaptation of the SPIN test. A similar
approach to the one used for the development of the
original version of the SPIN test was taken to establish the
Test de Phrases dans le Bruit (TPB). This paper describes
the development of the TPB, which consists of five French
lists of forty recorded-sentences and a speech babble.
The Development of the TPB
The approach used to develop the test lists of the
TPB involved the measurement of the intelligibility of the
key words in noise (Experiment 1), the evaluation of the
difference between the scores obtained on the HP and the
LP sentences (Experiment 2) and the verification of the
performance on the TPB at various SNRs (Experiment 3).
The series of experiments that lead to the development
of the preliminary version of the TPB is described below.
Development of the Speech Materials
According to Kalikow et al. (1977), to simplify the task
and to minimize the influence of linguistic and memory
skills, the type of response to be required from the subject
has to be a single word response. As for the SPIN test, it
was determined that the response word for the TPB would
be the last word of the sentence. This type of response is
also convenient for the examiner as the scoring simply
requires matching the response with the final word of the
test sentence (Kalikow et al., 1977). In order to further
control the linguistic content of the sentences, another
restriction was that the key word had to be a monosyllabic
word. Moreover, all the sentences were constrained to
contain six to eight syllables.
As opposed to the SPIN test, which was developed for
unilateral presentation of the sentences and the babble
262
Revue canadienne d’orthophonie et d’audiologie - Vol. 34, No 4, Hiver 2010
Development of the TPB
noise, a bilateral presentation mode was selected for the
TPB. This option was chosen based on the poor ecological
validity of unilateral presentation when testing speech in
noise (Besing, Koenke, Abouchacra, & Letowski, 1998;
Jerger, Greenwald, Wamback, Seipel, & Moncrieff, 2000).
Because the familiarity of the words influence their
intelligibility when they are presented in noise (Elliott et al.,
1979; Epstein, Giolas & Owens, 1968; Kalikow et al., 1977),
all the key words chosen for the test material were selected
from the MANULEX database (Lété, Sprenger-Charolles,
& Colé, 2004). The MANULEX is a web database listing
word frequency values for 48,886 lexical entries encountered
in 54 French books used in European French elementary
schools (Lété et al., 2004). No such large database was
available for words used in Canadian French children
literature. Monosyllabic words with a frequency of use
within the range of 7.7 to 935.4 per million words were
selected from the MANULEX. The initial pool consisted
of 200 key words. Within the constraints previously noted,
a set of 200 HP sentences was developed (e.g., Elle met la
nappe sur la table), as well as a set of 200 LP sentences (e.g.:
J’ai acheté une nouvelle table).
The resulting corpus of 400 sentences was analyzed
by two grade 3 teachers (i.e., teaching children of eight to
nine years of age), who were speakers of Canadian French,
to confirm the naturalness of the sentences. The teachers
were also invited to provide suggestions to improve the
naturalness of the sentence where needed. They were
asked to take into account that the TPB was to be used
with children and adults.
Following the revision of the sentence naturalness,
nine female native Canadian French speakers aged from 9
to 11 years completed a paper-and-pencil test to confirm
the predictability of the sentences. The 400 sentences
were listed on answer sheets with the key word deleted.
Participants were instructed to fill in the blank with a word
that they thought would most likely occur at the end. For
each of the HP sentences, if none of the participants had
written the intended key word, the sentence was reworked
to be more predictable. For each of the LP sentence, if one
participant had written the intended key word, the sentence
was reworked to be less predictable.
It was determined that the sentences should be recorded
by a female speaker because of the predominance of female
educators and caregivers in children’s education (Fallon,
Trehub & Schneider, 2000). A female speaker of Canadian
French who had previously participated in similar recording
sessions was chosen to produce the 400 revised sentences.
The sentences were recorded in a quiet recording room at
the University of Montreal, with a digital video camcorder
(Canon GL2, Canon Canada, Mississauga, ON L5T1P7)
to which an external lapel microphone (Audiotechnica
Pro70, Tokyo, Japan) was connected. During the recording
session, the camera was positioned at approximately
2.5 meters in front of the speaker. The microphone was
hanging from the ceiling, positioned at approximately 0.5
meters in front of the speaker. The speaker was instructed
to articulate each sentence as naturally and as clearly as
possible. The recordings were then organized into 400
individual sentence files using the iMovie 4 software (Apple
Canada, Markham, ON L3R 5G2). To ensure a uniform
level across the stimuli, the key words were edited with the
Cool Edit Pro software (Cool Edit Pro version 2.1, Adobe
Systems Canada, Toronto, ON M8X 2X3) to be within ±
2 dB of the root mean square average level (68.3 dB SPL)
of the 400 key words.
Since the key words were selected from the European
French MANULEX database, a test of word familiarity was
conducted with a group of children who were speakers of
Canadian French. This verification was necessary because
of the cultural differences between European and Canadian
French. Five lists of 40 key words were developed for the
familiarity test. The key words were all taken from the
recorded LP sentences audio files to ensure, as much as
possible, a similar accentuation on each word. The five lists
of key words were burned to individual audio compact discs
(CD). Forty children (19 girls and 21 boys) ranging from
5.5 to 7.4 years of age (average of 6.5 years) participated in
this study. A parent of each participant signed the consent
form and completed a questionnaire. Each participant was
tested individually in a quiet room where ambient noise level
did not exceed the specifications for hearing screening in
schools (ASHA, 1997). A hearing screening at the intensity
level of 20 dB hearing level (HL) was performed with a
portable audiometer (Maico MA 41, Maico GmbH, 10587
Berlin, Germany; Beltone AE2, Beltone, Glenview IL 60026)
with TDH-39 headphones (Telephonics, Farmingdale,
NY 11735) prior to the experiment. All participants had
normal hearing sensitivity at 500, 1000, 2000 and 4000
Hz bilaterally. The exclusion criteria for this study were
any history of language disorders, otological problems,
attention disorders or general learning delays. Four lists
were presented monaurally to each participant (two lists
per ear) via a CD player (Panasonic RX-D27, Mississauga,
ON, L4W 2T3) connected to the portable audiometer set
at 60 dB HL and a pair of headphones. The listener was
instructed to report each word that was presented and
to guess if necessary. A total of 160 words out of the 200
were correctly identified by over 80% of the participants.
This suggested that the majority of the selected words
were familiar to Canadian French children of five to seven
years old.
Following the familiarity testing of the key words,
60 words were removed from the corpus on the basis of
different considerations: (a) words with a frequency of
use score of less than 10 per millions words (according
to the MANULEX database) yielding a recognition score
of less than 50%, (b) homonymous words like boue and
bout and (c) words with different pronunciation across
Canadian French communities (e.g., zoo, oeuf, clown).
This eliminated 120 sentences from the pool of recorded
sentences because each key word appeared once in a HP
and once in a LP sentence.
The remaining 280 sentences were divided into seven
lists of 40 sentences, ensuring that the familiarity value
263 Canadian Journal of Speech-Language Pathology and Audiology - Vol. 34, No. 4, winter 2010
Development of the TPB
100
Procedure
90
Percentage of correct word recognition
Each participant was tested individually in
an audiometric suite, using the same audiometer
80
and headphones as for the hearing screening.
The sentences were transmitted via one CD
70
player (Panasonic RX-D27) connected to the
60
audiometer. The speech babble was conveyed via
another CD player (TASCAM CD-A500, TEAC
50
Canada, Mississauga, ON L4Z 1Z8) connected to
40
a different audio-input channel of the audiometer.
The seven lists of 40 sentences were presented
30
at a SNR of 0 dB (the sentences and the speech
babble at 65 dB HL) with monaural right ear
20
presentation. The selection of the SNR of 0 dB
10
was based on Kalikow et al.’s (1977) work for
the SPIN test. The speech babble of European
0
French talkers (4 females and 4 males) by Perrin
1
2
3
4
5
6
7
and Grimault (2005) was used. Among the
available pre-recorded babble, this was the most
Lists
representative of the babble conditions of the
Figure 1:
target population (i.e., speakers of Canadian
Group mean percent correct scores obtained by 10 adults for seven
French). The speech babble was recorded in a
lists of sentences at a signal-to-noise ratio of 0 dB (Experiment 1).
continuous loop on a separate CD.
For each list, the dark grey bar represents the word correct score for
The order of the lists of sentences was
the 20 high predictability sentences and the grey bar represents the
partially counterbalanced across the participants
word correct score for the 20 low predictability sentences.
(based on a Latin Square design). Participants
were instructed to report the last word of each sentence
of the key words was evenly distributed across the lists.
they heard and to guess if necessary.
Each list contained 20 HP and 20 LP sentences. A key
word appeared only once in a given list, as in the SPIN
Results
test (Kalikow et al., 1977). The lists were transferred onto
The percent correct score average for the HP and LP
seven separate CDs for the speech intelligibility in noise
items
and standard deviations for each list are provided in
testing described in the following section.
Figure 1. Across the seven lists, the word recognition score
ranged from 77% to 90.5% for the HP sentences (range
Experiment 1 - Measurement of the Key
of 13.5%), and from 58% to 74.5% for the LP sentences
Words’ Intelligibility in Noise
(range of 16.5%).
The goal of Experiment 1 was to determine if the speech
As for all the statistical analyses presented in this paper,
intelligibility in noise of the key words was homogeneous
an
arcsine
transform was applied to the data to stabilize
across the seven sentence lists.
the error of variance (Studebaker, 1985). An alpha level
of 0.05 was used for all the statistical comparisons unless
Participants
otherwise indicated. A repeated-measure, two-way analysis
Ten Canadian French speaking adults (five females
of variance (ANOVA) was performed on the mean average
and five males) between 19 and 28 years of age (average
score obtained for the HP sentences and the LP sentences
of 22 years) were recruited for the measurement of the key
at each list. The ANOVA was conducted with the factor
words’ intelligibility. Once the consent form was signed,
Type of sentences (HP and LP sentences) and the factor
each participant completed a questionnaire to rule out any
List (seven lists). There was a significant main effect of
exclusion criteria such as history of otological problems,
Type of sentence [F(1,9)= 98.73, p<.001, η2= 0.92] across
language delay, attention disorders or general learning
the seven lists. There was also a significant main effect of
delay. If none of the exclusion criteria were identified,
List [F(3.49,31.38)= 5.28, p<.001, η2= 0.37]. The interaction
the participants were asked to undergo a bilateral hearing
of Type of sentences x List was significant [F(6,54)= 2.73,
screening at 500, 1000, 2000 and 4000 Hz in an audiometric
p= .022, η2= 0.23]. This significant interaction was
test suite. Using a Midimate 622 audiometer (GN
anticipated given that the HP-LP difference score ranged
Otometrics, Schaumburg, IL 60173 5329), the test tones
appreciably across the lists, e.g., from 9% to 28.5%. Because
were presented at 15 dB HL with TDH 39 headphones. If
the sentence sets had to be re-worked to ensure an even
no sign of hearing loss was identified, the individual was
distribution of the key words’ intelligibility in noise values
invited to participate in the experiment.
across the seven lists, no further statistical analyses were
undertaken.
The sentences were re-assembled into a different set of
264
Revue canadienne d’orthophonie et d’audiologie - Vol. 34, No 4, Hiver 2010
Development of the TPB
Percentage of correct word recognition
100
Participants
90
A sample of 14 adults (11 females and
3 males) between 21 to 27 years of age
(average of 23 years), speakers of Canadian
French, was recruited for this study. None
of the participants had taken part in
Experiment 1. Prior to the experiment,
participants were asked to sign the consent
form and to complete a questionnaire
to rule out any exclusion criteria. The
inclusion and exclusion criteria used to
recruit the participants were the same as
Experiment 1.
80
70
60
50
40
30
20
10
Procedure
0
Each participant was tested individually
in
an
audiometric suite with the same
Lists
equipment as in Experiment 1. The seven
revised lists of 40 sentences were presented
Figure 2:
Group mean percent correct scores obtained by 14 adults for seven lists of sentences at a SNR of -2 dB (sentences at 65 dB HL and
at an SNR of -2 dB (Experiment 2). For each list, the dark grey bar represents speech babble at 67 dB HL) with monaural
the word correct score for the 20 HP sentences, the grey bar represents the word right ear presentation. The selection of
correct score for 20 LP sentences and the white bar represents the mean of the the SNR of -2 dB was based on pilot data
difference scores between the HP and the LP sentences.
obtained from three participants. The
pilot data indicated that the maximum
difference in performance between the HP and the LP
seven lists of 40 sentences, ensuring an even distribution
sentences was within that range of SNR. The same speech
of the key words’ score across the lists according to their
babble CD by Perrin and Grimault (2005), which was used
intelligibility values. Across the seven revised lists, the word
recognition score ranged from 83% to 86.5% for the HP
in the Experiment 1, was employed for this experiment. The
sentences (range of 3.5%), and from 67.5% to 65.5% for the
order in which the sentence lists were presented was partially
LP sentences (range of 2%). To ensure that the revised lists
counterbalanced across the participants (based on a Latin
were homogeneous, an ANOVA was performed on the mean
Square design). Participants were instructed to report the last
average score of the HP sentences and the LP sentences of the
word of each sentence they heard and to guess if necessary.
revised lists. The results revealed no significant main effect
Results
of List [F(6,114)= .08, p= 0.998, η2= 0.00]. The interaction
The
percent
correct
score average for the HP and LP
of Type of sentences x List was also not significant
2
items,
as
well
as
the
difference
of scores between the HP
[F(6,114)= .02, p= 1.000, η = 0.00]. The revised lists were
and
the
LP
sentences,
for
each
list
are summarized in Figure
recorded on seven separate CDs for the evaluation of the
2.
Across
the
lists,
the
scores
ranged
from 57.5% to 63.9%
difference of scores between the HP and the LP sentences
for
the
HP
sentences
and
from
34.3%
to 45% for the LP
across the lists, described in the following section.
sentences. The average difference scores between the HP
and the LP sentences ranged from 15% to 27%.
Experiment 2 –
A repeated-measures two-way ANOVA was performed
Evaluation of the Difference of the Scores
on the mean average score obtained for the HP sentences and
between the HP and the LP Sentences
the LP sentences for each list. The ANOVA was conducted
As the aim of the TPB is to provide a mean for evaluating
with the factor Type of sentences (HP and LP sentences)
the extent to which the listeners can take advantage of the
and the factor List (seven lists). There was a significant
linguistic context, the lists had to be equivalent not only
main effect of the Type of sentences [F(1,13)= 11.72, p< .001,
for the intelligibility in noise of the key words, but also
2
= 0.47], but the main effect of List did not reach
η
for the difference of scores between the HP and the LP
signifi
cance level [F(6,78)= 0.77, p= .60, η2= 0.06]. The
sentences. The goal of the Experiment 2 was to verify the
interaction of Type of sentences x List was significant
equivalence of the seven revised lists and to ensure that the
[F(6,78)= 29.3. p< 0.001, η2= 0.69], indicating that the
difference of scores between the HP and the LP sentences
difference score between the two types of sentences was
was homogenous across the lists.
influenced by the list. The results of these analyses suggested
that the lists were equivalent if considering the total
average of correct recognition score (HP and LP sentences
collapsed). However, when evaluating the average score
obtained for the HP and the LP sentences separately, the
1
2
3
4
5
6
7
265 Canadian Journal of Speech-Language Pathology and Audiology - Vol. 34, No. 4, winter 2010
Development of the TPB
Table 1
Samples of the actual version of the TPB sentence lists. The type of sentences is indicated in parentheses at the end of
each item, i.e., HP for the high predictable sentences and LP for the low predictable sentences. Each key word appears
once in HP and once in a LP sentence, but only once in a given list, for example, the word “camp” appears in the list 1
in the HP context (bold) and in the list 2, in the LP context.
Liste 1
1. Ce marchand vend des perles. (LP)
2. Claudie a découvert une mine. (LP)
3. Mon grand-père se berce sur sa chaise. (HP)
4. J’ai lu le livre jusqu’à la fin. (HP)
5. Il grave son nom sur du bronze. (LP)
6. Nos poumons respirent toujours de l’air. (HP)
7. Ma grand-mère a cousu ma robe. (LP)
8. J’ai quatre as dans mon jeu de cartes. (HP)
9. Jeanne se coupe les ongles. (LP)
10. Le chanteur a une très belle voix. (HP)
11. Tu as attaché ta tuque. (LP)
12. Elle lui fait signe de la main. (HP)
13. Certains soldats deviennent des fous. (LP)
14. Cette couverture est faite en laine.(LP)
15. J’enlève la neige avec une pelle. (HP)
16. Une main a quatre doigts et un pouce. (HP)
17. Mes enfants jouent avec une toile. (LP)
18. Ce cheval appartient au roi. (LP)
19. Ce joueur d’hockey fait des belles passes. (HP)
20. Le ballon roule vers le but. (LP)
Liste 2
1. Ce quilleur fait tomber toutes les quilles. (HP)
2. Jacinthe s’en va à son cours d’art. (LP)
3. Ils sont tous partis au camp. (LP)
statistical analyses revealed an effect of the list. This was an
indication that the average of the difference scores between
the HP and the LP sentences was not equivalent across the
sentence sets. Repeated-measures one-way ANOVA was
performed on the mean average of the difference scores
(between the HP and the LP sentences) obtained for each
list. The effect of list did not reach significance level [F(6,
= 1.98, p= .08, η2= 11.87]. However, it was felt that a
78)
revision of the lists was necessary because of the range
of the difference scores average across the lists, i.e., from
15% to 27%.
To obtain equivalent list sets and to maximize the
difference scores between the HP and the LP sentences,
individual analysis of each pair of sentences was performed.
The percentage value of the recognition score obtained for
the HP and LP sentences of each key word was compared.
For some key words, the percentage value obtained for
the HP and the LP sentence was similar. In other cases,
the percentage value of the recognition score for the LP
sentence was higher than the HP sentence (for the same
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
Ma cousine a trouvé un gros os. (LP)
Cette chanson s’appelle « vive le vent ». (HP)
J’ai acheté de la gomme. (LP)
On voit mieux la lune pendant la nuit. (HP)
Les deux garçons jouent à la guerre. (LP)
Il est mort quand j’avais cinq ans. (HP)
Ils chantent autour du feu de camp. (HP)
Les deux amis ont fait la paix. (LP)
Il faudra mettre une deuxième couche. (LP)
J’enferme mon chat dans sa cage. (HP)
Tous les trains roulent sur des rails. (HP)
Nous lui avons donné un verre. (LP)
Ce chandail n’a pas de prix. (LP)
Il fend le bois avec une hache. (HP)
Le frappeur a frappé la balle. (HP)
Maman a coupé les fleurs. (LP)
L’avion vole haut dans le ciel. (HP)
Le fermier va nourrir ces vaches. (LP)
Le jardinier arrose ses plantes. (HP)
Les girafes ont un grand cou. (HP)
4. J’aime le beurre à l’ail. (HP)
5. …
key word). In both instances, the pairs of sentences had to
be eliminated from the corpus as the constraints were not
met. A total of 80 sentences were removed from the corpus
(i.e., 40 key words), based on these individual analyses.
The remaining 100 HP sentences and 100 LP sentences
were assembled into five lists of 40 sentences, ensuring
an even distribution of the key words according to
their familiarity and intelligibility in noise values (from
Experiment 1). Precautions were also taken for having
equivalent means of the difference scores between the HP
and LP sentences across the lists (from the results obtained in
Experiment 2). The five lists were recorded on one CD, each
list on a different track. This constituted the preliminary
version of the sentence lists set of the TPB (see a sample
of the TPB lists in Table 1). The performance at various
SNRs had to be verified. This verification is described in
the following section.
266
Revue canadienne d’orthophonie et d’audiologie - Vol. 34, No 4, Hiver 2010
Development of the TPB
100
Percentage of correct word recognition
90
**
**
80
70
Figure 3 (left):
Group mean percent correct scores obtained
by 15 adults at various SNRs with the TPB
sentence lists (Experiment 3). The solid line
illustrates the performance with the HP
sentences and the broken line illustrates the
performance with the LP sentences at each
SNR. Significant differences between both
types of sentence are indexed with stars (*
p = 0.01; ** p < 0.001).
**
60
50
40
**
30
20
*
10
0
-6
-4
-2
0
+2
Signal-to-noise ratio (dB)
score
Percent transforms
Figure 4 (below):
The group mean percent correct scores illustrated in Figure 3 were transformed into z-scores. The z-scores as a function of the
SNR for the HP sentences (square symbol) and the LP sentences (diamond symbol) are illustrated in the left panel. The linear
regression function derived from the z-scores is illustrated with a solid line for the HP sentences and a broken line for the LP
sentences. In the right panel, an Ogive (cumulative frequency) plot of target word intelligibility is shown, with a solid line for the
HP sentences and a broken line for the LP sentences. The SNR at which a 50% key word intelligibility score would be reached is
indicated by the symbol « x ».
Signal-to-noise ratio (dB)
Experiment 3 Verification of the Performance on
the TPB at Various SNRs
The objective of this experiment was to verify the
performance on each type of sentence as a function of SNR.
Participants
A group of 22 Canadian French speaking adults was
recruited for this study. None of them had taken part in
Experiment 1 or 2. The age range extended from 19 to 43
Signal-to-noise ratio (dB)
years (average of 27 years). As for the previous experiments,
all the participants were required to sign a consent form
and to complete a questionnaire to rule out the presence of
any exclusion criteria. The inclusion and exclusion criteria
were the same as those used for the Experiment 1.
Seven participants had to be excluded from the
study. Two participants reported a diagnosis of attention
disorder during their childhood. One participant failed
the audiological screening assessment. The data from four
participants were discarded because they only completed
267 Canadian Journal of Speech-Language Pathology and Audiology - Vol. 34, No. 4, winter 2010
Development of the TPB
50%
HP-LP difference score
40%
30%
20%
10%
0%
-8
-6
-4
-2
0
2
4
Signal-to-noise ratio (dB)
Figure 5:
Difference of scores between the HP sentences and the LP
sentences (in percent) as a function of signal-to-noise ratio
(SNR), computed from the ogive functions illustrated in
Figure 4.
four out of the five experimental conditions (due to lack of
time). In total, the data from 15 participants (nine females
and 6 males) were included in the analyses.
Procedure
This experiment was conducted in a quiet room at the
Université du Québec à Trois-Rivières (Québec, Canada),
which met the ANSI S3.1-1999 (R2008) specifications. The
same equipment used for Experiment 1 and 2 was employed,
i.e., Midimate 622 audiometer and TDH 39 headphones.
The audiometer was connected to two Panasonic RX-D27
CD players. For this experiment, the testing conditions were
similar to those of the TPB, i.e., bilateral presentation of
the sentences and the noise. A local audiometric equipment
company (Genie Audio Inc., Saint-Laurent, QC H4N
1T1) was consulted and developed an audio mixer that
allowed bilateral presentation of both the test stimuli and
the masking noise as well as an independent control of the
intensity level of each one.
Each participant was tested individually with the five
lists of the TPB and the same speech babble (Perrin &
Grimault, 2005) as for the previous experiments. All the
participants listened to each of the five lists presented
at five different SNRs, i.e.: -6, -4, -2, 0 and +2 dB. The
sentences lists were always presented at 60 dB HL. The
order of presentation of the lists and SNRs was partially
counterbalanced across the participants (based on a Latin
Square design). The sentences and the babble noise were
presented bilaterally.
Results
Mean results for the experiment are summarized in
Figure 3. The percent correct word recognition scores for
the HP and LP sentences obtained at each SNR are provided.
The most consistent finding was that the mean average of
the correct scores for the HP sentences was higher than
for the LP sentences percentage at the five tested SNRs. A
repeated-measures two-way ANOVA was performed on
the mean average score obtained for the HP and the LP
sentences at each SNR to test the statistical significance
of these trends. The analysis of variance was conducted
with two within-subject factors: Type of sentences (HP
and LP sentences) and SNR (five levels corresponding to
the five tested SNRs). The analysis revealed a significant
main effect of Type of sentences [F (1,56) = 268.35,
p< .001, η2= 0.95] and a significant main effect of SNR
[F(4,56) = 273.97, p< .001, η2= 0.95]. The interaction
of the Type of sentence x SNR was also significant
[F(2.3,56) = 8.46, p< .001, η2= 0.38], suggesting that the
difference score between the two types of sentences was
influenced by the SNR. This was probably caused by the floor
and ceiling effects of the performance-intensity function.
For example, at -6 dB SNR, the performance for both LP
and HP items approached 0%, reducing the difference of
score between the two types of sentences.
Additional analyses were conducted to explore the
nature of the Type of sentence x SNR interaction. This
was accomplished by comparing the performance of the
HP and the LP sentences at each SNR. Five paired t-tests
indicated that the performance for the HP sentences were
significantly different from the performance obtained with
the LP sentences at each of the five SNRs tested, using the
Bonferroni correction (critical alpha level of 0.01).
As in other studies on speech recognition performance
(Boothroyd & Nittrouer, 1988; Laroche et al., 2003; Mayo
et al., 1997), the mean percent correct scores for each type
of sentences at each SNR was transformed into z scores.
The z scores as a function of the SNR for each type of
sentence is illustrated in the left panel of Figure 4. A linear
regression function was calculated with the z scores. As
shown in Figure 4, the data were well fitted by the linear
regression function. The r2 variance accounted for was
over 0.9. The functions obtained for the HP and the LP
sentences roughly showed a similar slope, i.e., 0.381 z/dB
for the HP sentences and 0.400 z/dB for the LP sentences.
From the linear regression function, the z-scores were
converted back to percentages to produce the intelligibility
ogive (cumulative frequency) plots for the LP and HP
sentences, as shown in the right panel of Figure 4. The
data obtained with the HP and LP sentences of the TPB
provided typical ogive speech intelligibility functions as
the SNR increased. Based on these functions, it is noted
that 50% key word intelligibility is reached at a lower
SNR with the HP sentences (-2.8 dB) than with the LP
sentences (-0.85 dB). This difference of SNR illustrates
the contribution of the linguistic contextual information
to auditory speech perception.
Difference scores were used in other studies to
characterize the gain in speech perception performance
attributable to the provision of additional linguistic and
contextual cues (Elliott & Busse, 1987; Erber, 1975; PichoraFuller, 2008; Pichora-Fuller et al., 1995) or by the provision
of visual information (Gagné, Tugby, & Michaud, 1991;
Ross, Saint-Amour, Leavitt, Javitt, & Foxe, 2007). Therefore,
268
Revue canadienne d’orthophonie et d’audiologie - Vol. 34, No 4, Hiver 2010
Development of the TPB
an HP-LP sentences difference score was computed from
the ogive functions illustrated on the right panel of Figure
4. The plot of the HP-LP difference of scores as a function
of SNR is illustrated in Figure 5. The plot reveals an inverted
u-shaped relationship between the gain in recognition
accuracy due to the HP sentences’ additional linguistic
contextual cues and the SNR. Using the difference score
measure, it appears that the maximum benefit of the
linguistic contextual cues for this group of listeners occurs
at the center of the curve, at the SNR of -1.5 dB, with a gain
of 25%. This observation is attributable to the particular
test conditions used in the present experiment.
Discussion
This paper described the development of the TPB,
which is a French adaptation of the SPIN test. SPIN-like
tests provide a useful and time-tested way to measure speech
recognition performance in the presence of background
noise. The results obtained from the TPB may be analyzed
from different perspectives, for example, by studying the
difference between the scores on HP and LP sentences
as a function of SNR. This perspective illustrates the
contribution of language knowledge and ability to use
the linguistic context of the HP sentences to understand
speech (Elliott & Busse, 1987). It also shows at which
SNR the listener benefits the most from the linguistic and
contextual cues (Pichora-Fuller, 2008; Pichora-Fuller et
al., 1995). For the group of adults who participated in this
study, it appears that the maximum benefit of the linguistic
contextual cues occurs at the SNR of -1.5 dB, with a gain
of 25%. This observation is limited to the particular test
conditions of the present study. However, in the case of
listeners with hearing problems, the maximal difference
score may fall at a different SNR because of the shift in
the listener’s performance-SNR curve and the possible
difference in slope of the LP and HP curves. Moreover, in
the case of listeners who cannot benefit from the linguistic
contextual cues because of a language deficit, the magnitude
of the difference scores between the HP and LP sentences
may be lower.
The exploration of the speech perception problems
experienced by individuals with auditory processing
disorder (APD) counts among the applications of the
TPB. The American Speech-Language-Hearing Association
(2005) describes APD as difficulties in the perceptual
processing of auditory information at the level of the
central auditory nervous system. However, at the present
time, the results of available studies have not specifically
and unequivocally identified the underlying causes of the
reported speech perception problems in noise reported
by individuals with APD. If the underlying dysfunction
in the case of APD is related to the auditory processing
of the acoustic speech signal and not to the languagebased processing, listeners with APD should be equally
competent at using linguistic contextual cues at the TPB
as individuals without listening problems. At present,
many general intervention programs proposed for the
rehabilitation and management of APD include procedures
to increase auditory closure abilities in order to improve
the use of linguistic contextual information to facilitate
speech perception in noise. Auditory closure refers to
the recognition of complete words, or utterances, when
only parts are spoken or heard (Delk, 1991). However,
if the TPB results demonstrate that listeners with APD
have similar auditory closure abilities as control groups,
such intervention may not be required. Findings of this
sort would guide the professionals involved with listeners
presenting with APD to develop more effective intervention
plans.
The TPB could also be used with other populations
with speech perception problems in noise, to investigate
their auditory closure skills which is not possible with
other available French speech in noise tests. However, more
testing with these clinical populations will be necessary
to determine the diagnostic properties and accuracy of
the TPB.
Additional evaluations of the psychometric properties
and diagnostic usefulness of the TPB must be performed
before it can be routinely applied in research and clinical
applications. First, the equivalencies of the actual lists have
to be measured in more detail. Second, in order to ensure
that the TPB is appropriate for children, the test will have
to be evaluated with that population. Normative data will
also have to be collected for both the adult and children
populations before its routine use, to allow comparison
of performance measured with populations presenting
with speech perception problems in noise. Moreover, for
the data collection, additional validation for different
dialects of French (other than Canadian French) may
have to be undertaken, as the performance on the TPB
may be influenced by the dialect, like any other speech
perception test.
Conclusion
The objective of this paper was to describe the initial
steps used to develop the TPB. The present findings should
be interpreted with caution as only individuals with
normal hearing function participated in the experiments.
Additional evaluations of the psychometric properties of
the test have to be performed before its clinical applications.
Nevertheless, the preliminary findings suggest that further
development of the TPB is warranted. The sentence lists that
resulted from the research described here will be useful for
the exploration of the underlying auditory and/or languagebased origins of speech perception problems in noise for
speakers of Canadian French. A better understanding of the
perception of speech in noise may inform the development
of more specific and effective intervention programs.
Acknowledgments
The authors wish to thank Andréa Bissonnette, Amélie
Gaudreault, Mélanie Gagnon, Charlotte Ballet, Marie-Josée
Levasseur, Gassia Jakmakjian, Yang Huang and MarieClaude Chouinard for their assistance at various stages
of the data collection. A special thanks to Anne-Marie
Hurteau for agreeing to be the talker in our recordings,
269 Canadian Journal of Speech-Language Pathology and Audiology - Vol. 34, No. 4, winter 2010
Development of the TPB
as well as to all the participants at the various stage of the
test development. The authors would also like to extend
their thanks to the Conseil des Écoles Publiques de l’Est de
l’Ontario and the Université du Québec à Trois-Rivières
for their collaboration to this project.
Portions of this paper were presented at the 9th Congress
of International Commission on Biological Effect of Noise in
Mashantucket (July 2008) and at the Colloque international
de réadapation sur la surdité, la surdicécité et les troubles du
langage et de l’audition in Montréal (June 2009).
This work was supported by a doctoral fellowship from
the Centre de recherche du CHU Sainte-Justine and Fonds
québécois de la recherche sur la nature et les technologies.
References
American Speech-Language-Hearing Association. (1997). Guidelines for
Audiologic Screening [Guidelines]. Retrived on March 14, 2010 from www.asha.
org/policy.
American Speech-Language-Hearing Association. (2005). (Central)
Auditory Processing Disorders [Technical Report]. Retrieved on March 14, 2010
from http://www.asha.org/members/deskref-journals/deskref/default.
ANSI S3.1. 1999 (R2008).Maximum Permissible Ambient Noise Levels
for Audiometric Test Rooms ANSI S3.1. 1999. American National Standards
Institute, New York.
Besing, J., Koenke, J., Abouchacra, K., & Letowski, T. (1998). Contemporary
approaches to audiological assessment in young children. Topics in Language
Disorders, 18, 52-70.
Bilger, R.C., Nuetzel, M.J., Rabinowitz, W.M. & Rzeczkowski, C. (1984).
Standardization of a test of speech perception in noise. Journal of Speech, Language,
and Hearing Research, 27, 32-48.
Boothroyd, A., & Nittrouer, S. (1988). Mathematical treatment of context
effects in phoneme and word recognition. Journal of the Acoustical Society of
America, 84, 101-114.
Bradlow, A.R., & Alexander, J.A. (2007). Semantic and phonetic
enhancements for speech-in-noise recognition by native and non-native listeners.
Journal of the Acoustical Society of America, 121, 2239-2349.
Delk, J.H. (1991). Comprehensive dictionary of audiology (4th edition).
Maynards: The Laux Company.
Dubno, J.R., Ahlstrom, J.B., & Horwitz, A.R. (2000). Use of context by
young and aged adults with normal hearing. Journal of the Acoustical Society of
America, 107, 538-546.
Elliott, L.L. (1995). Verbal auditory closure and the speech perception
in noise (SPIN) Test. Journal of Speech, Language, and Hearing Research, 38,
1363-1376.
Elliott, L.L., & Busse, L.A. (1987). Auditory Processing by Learning Disabled
Adults. In D. Johnson & J. Blalock (Eds.), Adults with Learning Disabilities: Clinical
studies (pp.107-129). New York: Grune & Stratton.
Elliott, L.L., Connors, S., Kille, E., Levin, S., Ball, K., & Katz, D. (1979).
Children’s understanding of monosyllabic nouns in quiet and in noise. Journal
of the Acoustical Society of America, 66, 12-21.
Epstein, A., Giolas, T.G., & Owens, E. (1968). Familiarity and Intelligibility
of Monosyllabic Word Lists. Journal of Speech, Language and Hearing Research,
11, 435-438.
Erber, N.P. (1975). Auditory-visual perception in speech. Journal of Speech
and Hearing Disorders, 40, 481-492.
Fallon, M., Trehub, S.E. & Schneider, B.A. (2000). Children’s perception
of speech in multitalker babble. Journal of the Acoustical Society of America,
108, 3023-3029.
Florentine, M. (1985). Speech perception in noise by fluent, non-native
listeners. Journal of the Acoustical Society of America, 77, S106-S106.
Gagné, J.P., Tugby, K.G., & Michaud, J. (1991). Development of a
Speechreading Test on the Utilization of Contextual Cues (STUCC): Preliminary
Findings with Normal-Hearing Subjects. Journal of the Academy of Rehabilitative
Audiology, 24, 157-170.
Humes, L.E., Burk, M.H., Coughlin, M.P., Busey, T.A., & Strauser, L.E.
(2007). Auditory Speech Recognition in Younger and Older Adults: Similarities
and Differences Between Modalities and the Effects of Presentation Rate. Journal
of Speech, Language, and Hearing Research, 50, 283-303.
Jerger, J., Greenwald, R., Wamback, I., Seipel, A., Moncrieff, D. (2000).
Toward a more ecologically valid measure of speech understanding in background
noise. Journal of the American Academy of Audiology, 11, 273-282.
Kalikow, D.N., Stevens, K.N., & Elliott, L.L. (1977). Development of a test of
speech intelligibility in noise using materials with controlled word predictability.
Journal of the Acoustical Society of America, 61, 1337-1351.
Laroche, C., Soli, S., Giguère, C., Lagacé, J., Vaillancourt, V., & Fortin, M.
(2003). An Approach to the Development of Hearing Standards for HearingCritical Jobs. Noise and Health, 6, 17-37.
Lété, B., Sprenger-Charolles, L., & Colé, P. (2004). MANULEX : A gradelevel lexical database from French elementary-school readers. Behavior Research
Methods, 36, 156-166.
Mayo, L.H., Florentine, M., & Buus, S. (1997). Age of Second-Language
Acquisition and Perception of Speech in Noise. Journal of Speech, Language and
Hearing Research, 40, 686-693.
Perrin, F. & Grimault, N. (2005). Fonds sonores. Laboratoire Unités Mixtes
de Recherche, Centre National de la Recherche Scientifique 5020, Lyon, France.
Pichora-Fuller, K.M. (2008). Use of supportive context by younger and
older adult listeners: Balancing bottom-up and top-down information processing.
International Journal of Audiology , 47, S72-S82.
Pichora-Fuller, K.M., Schneider, B., & Daneman, M. (1995). How young
and old adults listen to and remember speech in noise. Journal of the Acoustical
Society of America, 97, 593-608.
Ross, L.A., Saint-Amour, D., Leavitt, V.M., Javitt, D.C., & Foxe, J.J. (2007).
Do You See What I Am Saying? Exploring Visual Enhancement of Speech
Comprehension in Noisy Environments. Cerebral Cortex, 17, 1147-1153.
Schum, D.J., & Matthews, L.J. (1992). SPIN test Performance of Elderly
Hearing-Impaired Listeners. Journal of the Acoustical Society of America, 3, 303-307.
Studebaker, G.A. (1985). A “Rationalized” Arcsine Transform. Journal of
Speech, Language and Hearing Research, 28, 455-462.
Author’s Note
Correspondence concerning this article should be
addressed to Josée Lagacé, Audiology and Speech-Language
Pathology Program, University of Ottawa, Roger Guindon
Hall, 451 Smyth Road, Ottawa, Ontario, K1H 8M5. Email:
[email protected]
Received: October 1, 2009
Accepted: March 24, 2010
270
Revue canadienne d’orthophonie et d’audiologie - Vol. 34, No 4, Hiver 2010

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project