Original Paper Published online: November 13, 2015

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

Effects of Congenital Visual Deprivation on the Auditory Perception of Anticipatory Labial Coarticulation Lucie Ménard a Marie-Agnès Cathiard b Emilie Troille c Marilyn Giroux a  

a

 

 

 

Laboratoire de phonétique, Université du Québec à Montréal, Center for Research on Brain, Language, and Music, Montreal, Que., Canada; b Centre de Recherches sur l’Imaginaire EA 610, Université Stendhal, and c Troille Speech Language Pathology, Grenoble, France  

 

 

Abstract Objective: It has been shown previously that congenitally blind francophone adults had higher auditory discrimination scores than sighted adults. It is unclear, however, if, compared to their sighted peers, blind speakers display an increased ability to detect anticipatory acoustic cues. In this paper, this ability is investigated in both speaker groups. Methods: Using the gating paradigm, /izi/ and /izy/ sequences were truncated to include a variable duration of the vowel. The sequences were used as stimuli in an auditory identification test. Seventeen congenitally blind adults (9 females and 8 males) and 17 sighted controls were recruited. Their task was to identify the second vowel of the sequence. Results: Results show that all participants could reliably identify the rounded vowel prior to its acoustic onset, but steeper identification slopes were found for sighted listeners than for blind listeners. Conclusion: The difference in identification slopes likely suggests that sighted speakers display finer abilities to perceptually follow the decreasing values of the frication noise, compared to blind speakers. © 2015 S. Karger AG, Basel

© 2015 S. Karger AG, Basel 1021–7762/15/0672–0083$39.50/0 E-Mail [email protected] www.karger.com/fpl

Introduction

It is well known that speech gestures are highly coarticulated. Linguistic as well as perceptual-motor factors govern articulatory timing between gestures. For instance, during the production of a phoneme sequence, a given articulatory gesture associated with the second phoneme may be initiated well before the offset of the articulatory gestures associated with the first phoneme. Such a phenomenon is referred to as ‘anticipatory coarticulation’. At the perceptual level, studies have shown that anticipatory cues are functional in speech perception [1–3]. Indeed, they provide listeners with indices about the upcoming vowel. Several studies have focused on the perception of anticipatory coarticulation of lip protrusion and constriction gestures (lip rounding). Focusing on adult controls in the auditory modality and using the gating paradigm, researchers [4] tested the ability of French- and English-speaking listeners to identify the vowel V in [kstrV], [rstrV], and [rskrV] sequences, where V is one of the vowels /i/ (unrounded), /y/ (rounded), or /u/ (rounded). Several truncated versions of the sequences were presented, including gradually increasing vowel duration. Results showed that listeners could reliably perceive the vowel, in the auditory domain, well before the acoustic onset of that vowel. Similar results are found for Lucie Ménard Laboratoire de phonétique, Université du Québec à Montréal Center for Research on Brain, Language, and Music CP 8888, succ. Centre-Ville, Montreal, QC, H3C 3P8 (Canada) E-Mail menard.lucie @ uqam.ca

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

Key Words Anticipatory coarticulation · Sighted listeners · Blind listeners

84

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

in previous speech perception and production experiments conducted in our laboratory [10, 12, 13]. Agematched sighted adults were also tested.

Methods A perceptual identification test using the gating paradigm was conducted, in accordance with Troille et al. [14]. Stimuli A native adult male speaker of Quebec French was audio-visually recorded while uttering several repetitions of /izy/ sequences, at a normal speech rate. The sequence /izi/ was also recorded and served as a control condition. The acoustic signal was recorded with a high-quality microphone and digitized at 44,100 Hz. The signal was then low-pass filtered (with a cutoff frequency of 11,025 Hz) and downsampled at a frequency of 22,050 Hz. One repetition of each sequence was selected on the basis of signal quality. Lip area values, as calculated on the still image of the lips at vowel midpoint, were 1.4 and 1.3 cm2, respectively, for both vowels of /izi/, and lip area values were 1.2 and 0.4 cm2, respectively, for both vowels of /izy/. Those geometrical values are in line with those reported for French speakers [9, 14]. The fricative consonant /z/ was chosen because it is phonologically a nonlabial consonant, thus allowing the anticipatory rounded gesture associated with the phonologically labial /y/ vowel to be produced through this consonant. It has been reported that the labial vowel lengthens the front oral cavity, thus decreasing the frequency of the frication noise in /z/ [15–17]. The duration of the /z/ consonant was 192 ms. For each of the /izy/ and /izi/ sequences, 16 gated versions were obtained, with a 16-ms frame as the gating step. Each version had the same onset and included the first vowel /i/, but ended at different time points during the /zV/ part. First, the acoustic offset of the fricative consonant, corresponding to the moment where the vowel formants were appearing, was labeled. This point, which delimited the end of the first gated stimulus (labeled 0), served as a reference for the other truncation points. The 16-gated versions of each sequence ranged from the starting point of the sequence and included up to 3 frames after the fricative offset (stimuli 14, 15, and 16) and up to 12 frames before the fricative offset (stimuli 1–13). Figure 1 displays the acoustic signal for /izy/ and /izi/ as well as the overall time window of the 16-gated stimuli extracted from each of the original sequences. Each gated stimulus was repeated 6 times for a total of 192 stimuli (16 gates × 6 repetitions × 2 sequences). Acoustic analyses were performed on the /izy/ and /izi/ sequences using Praat software to relate the perceived scores with their underlying acoustic parameters. For each of the two vowels, formant values were extracted from a linear predictive coding (LPC) analysis, with a coefficient order of 24. For the /z/ consonant, two acoustic measures were used [18]. First, the center of gravity [19] was extracted for each nonoverlapping 10-ms hamming window of frication noise, in the 500- to 11,025-Hz range, after pre-emphasis of the signal (from 50 Hz) [16]. Second, to account for the fact that coronal fricatives exhibit formant structures, an LPC spectral analysis was performed using 10-ms nonoverlapping windows located at the end of each truncation point using a coefficient order of 24. Results of those analyses are displayed in figure 2 (dashed lines). The frequency of the second

Ménard/Cathiard/Troille/Giroux

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

French rounded vowels [5–7]. Together, those studies show that listeners can recover labial anticipatory cues provided by a speaker and use them to identify a vowel. In the visual modality, anticipatory lip rounding and constriction cues are also used to visually perceive an upcoming rounded vowel [3]. When presented with truncated stimuli (gating paradigm) in the visual-only condition, participants can identify the rounded vowel when the lip area decreased to approximately 1 cm2, before the change in formant structure associated with lip rounding [8, 9]. Thus, the varying shapes of the speaker’s lips provide the listener with anticipatory cues that are functional in the speech perception process. When comparing the results of identification scores in the auditory (audio-only) condition and in the audio-visual condition, it was found that for most subjects, the upcoming rounded vowel /V/ could be identified in a truncated /C-labialV+labial/ sequence earlier in the audio-only condition than in the audio-visual condition, in which stimuli are identified earlier than in the visual-only condition. Since visual and auditory consequences of anticipatory lip rounding are exploited in typical speech perception, it is particularly interesting to examine visually deprived listeners to investigate the importance of visual cues. In a previous study [10], we showed that congenitally blind francophone adults had higher perceptual scores in vowel discrimination tasks than sighted adults. Those results were obtained with synthesized steady-state vowels. It is unclear, however, if, compared to their sighted peers, blind speakers display an enhanced ability to detect anticipatory acoustic cues to rounding. It might be the case, considering the fact that congenitally blind speakers had to learn to efficiently process speech without being able to rely on complementary visual cues. A recent study [11] investigated the perception of anticipatory labial gestures in 10 blind and 10 sighted participants, using truncated versions of /isy/, /isø/, /isu/, and /iso/ sequences as stimuli in auditory identification tests. The results suggested that the blind group could identify the high rounded vowels /y/ and /u/ earlier in the fricative consonant than the sighted group. The mid vowels /ø/ and /o/ were not reliably identified before their acoustic onset. However, the study did not include any statistical analyses, and it is not clear whether participants in the blind group were congenitally blind or became visually impaired after birth. In the present paper, we investigate the effects of congenital visual deprivation on the perception of anticipatory labial coarticulation in French in the auditory modality, using the gating paradigm. Participants were Quebec French congenitally blind adults who had taken part

a

b

/izy/ 0.2475

/izi/

0.217 0

0

–0.2763

0.7034

–0.2991 8,000

0

Frequency (Hz)

Frequency (Hz)

8,000

0

0

0.7034

Time (s)

0

0

0.7663

0

0.7663

Time (s)

Gated zone

Gated zone 0.7034

0.7663

Time (s) Onset, stimuli 1–16

Offset, stimulus 1

Time (s) Onset, stimulus 16

Offset, stimulus 1

Onset, stimuli 1–16

Offset, stimulus 16

Fig. 1. Waveform and spectrogram of the /izy/ (a) and /izi/ (b) sequences used in the test. The two vertical dotted

/izy/ 8,000

£

b

££££££

Sighted Blind

£ 8,000

100

£ /izi/ 100

Sighted Blind

80

Center £of£ gravity£

60

40

4,000

Frequency (Hz)

6,000

% identification /y/

Frequency (Hz)

80 6,000

60

40 4,000

% identification /y/

a

Color version available online

lines delimit the gated zones (all stimuli ended at a time point located within that zone).

20

20 F2 2,000 –12

0 –7 –2 Position relative to /z/ offset

3

0

2,000 –12

–7 –2 Position relative to /z/ offset

3

Perception of Rounding in Blind Speakers

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

85

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

Fig. 2. Percent /y/ responses for blind (dotted line) and sighted (solid line) participants as a function of gate (0 corresponds to consonant offset or vowel acoustic onset) for the /izy/ (a) and /izi/ (b) sequences. Data are averaged across 17 participants in each group. Errors are standard errors. Values of /z/ center of gravity and F2 (in Hz) correspond to the dashed lines. The vertical arrow corresponds to the 50% category boundary.

Table 1. Characteristics of the 17 blind speakers

Subject

Gender Age, years Etiology of blindness

Vision at birth

Current vision

S1B

F

48

retinitis pigmentosa

U

S2B

F

40

congenital cataract

U

S3B S4B

F M

26 52

U optic atrophy

U total blindness

S5B

M

40

detachment of the retina

U

S6B

M

42

U

S7B

M

36

congenital cataract and congenital glaucoma retinitis pigmentosa

RE = 3/210 LE = 0 RE = 0 LE = 6/1,260 U (total blindness) RE = 0 LE = 0 RE = 2/180 LE = 2/105 U (total blindness)

total blindness

S8B

M

52

congenital cataract

total blindness

S9B

F

51

retinitis pigmentosa

total blindness

S10B S11B

F M

45 42

congenital cataract congenital glaucoma

total blindness U

S12B

F

45

congenital cataract

total blindness

S13B S14B

M M

40 47

retinitis pigmentosa congenital cataract

U total blindness

S15B S16B

F F

49 39

congenital cataract retinitis pigmentosa

U total blindness

S17B

F

41

retinitis pigmentosa

U

RE = 20/400 LE = 20/400 RE = 3/180 LE = 2/180 RE = 2/400 LE = 2/400 U (total blindness) RE = 2/180 LE = 3/180 RE = 0 LE = 0 U (total blindness) RE = 0 LE = 0 U (total blindness) RE = 3/400 LE = 2/400 U (total blindness)

LE = Left eye; RE = right eye; U = undetermined.

Participants and Experimental Procedure Seventeen congenitally blind adults (9 females and 8 males) were recruited. All subjects were native speakers of Canadian French living in the Montreal area. They had congenital, complete visual impairment, classified as class 3, 4, or 5 in the International Disease Classification of the World Health Organization. These 3 classes correspond to blindness. Table  1 summarizes the age, gender, and visual characteristics of the 17 blind subjects. The subjects had never had any perception of light, and they were 26–

86

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

52 years old (mean age, 43). They did not report any language disorders or motor deficits. A group of 17 age-matched native speakers of Quebec French who were recruited in the Montreal area served as control subjects (9 females and 8 males). They ranged in age from 25 to 50 years (mean age, 41). They all had normal vision or impaired vision corrected by lenses. They did not report any speech disorders or motor deficits. All subjects passed a 20-dB hearing level, pure-tone audiometric screening procedure at 500, 1,000, 2,000, 4,000, and 8,000 Hz. They all gave formal consent. The perception test was individually administered in a quiet room. Participants were seated comfortably in a chair in front of the experimenter. The 192 stimuli described above were presented in random order. The participant was told to identify aloud the second vowel in the sequence they heard among a choice of two French vowels: /i/ and /y/. The experimenter then entered the participant’s response by clicking on the appropriate vowel on the computer screen, which was not visible to the subjects.

Ménard/Cathiard/Troille/Giroux

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

spectral peak was measured on the LPC-based spectrum. The center of gravity of the spectrum varied from 6,942 to 7,625 Hz during the /izi/ sequence, while it decreased from 7,053 to 5,362 Hz during the /izy/ sequence. The second peak of the LPC analysis remained stable at around 4,025 Hz for the consonant in /izi/ and varied from 3,501 to 2,068 Hz during the /z/ consonant in /izy/. Thus, the anticipatory effects of the rounded /y/ were well produced compared to the /izi/ sequence. The rest of this paper focuses on the perceptual results of anticipatory lip rounding in /izy/.

Statistical Analyses To measure the effect of the speaker group, repeated-measures ANOVAs were conducted on the percent /y/ responses with the participant group (sighted or blind) as the between-subject factor and with gate (from 1 to 16) and sequence (/izi/ or /izy/) as the within-subject factors. Category boundaries and slopes of the identification functions were evaluated through Probit modeling [14, 20].

i/y identification slope –0.6

–0.5

–0.4

–0.3

–0.2

–0.1

Identification curves were determined, with 50% identification boundaries and identification slopes. Acoustic parameters were then compared to the identified results. Identification Curves Figure 2 shows the percentage of correct /y/ responses as a function of the truncation point of the sequence relative to the fricative offset, for blind (dotted line) and sighted participants (solid line). The scores are presented for the /izy/ sequence on the left panel and for the /izi/ sequence on the right panel. Error bars are standard errors. Data were averaged across speakers. First, it can be seen that the identification curves substantially differ between both sequences, and that the presence or absence of an upcoming rounded vowel affects identification scores. Between-speaker standard deviation of the percentage of correct /i/ and /y/ responses ranged from 0 to 22.8% (as can be observed in fig. 2), which is in the range of the standard deviations reported by Troille et al. [14]. Repeatedmeasures ANOVAs conducted on the percent scores with the speaker group as the between-subject factor and gate and sequence as the within-subject factors revealed a significant effect of sequence on the perceptual scores [F(1, 32) = 68,64; p < 0.001]. In the remaining part of the paper, the /izy/ sequence will be analyzed. Figure 2 also shows that the 50% threshold on the averaged identification curves, at which at least 50% of the tokens were correctly identified as /y/, occurred at gate –6.46 for the sighted participants and at gate –6.02 for the blind participants. Those time points correspond to –103.36 ms (for the sighted participants) and –96.32 ms (for the blind participants) before the rounded vowel’s acoustic onset. Thus, anticipatory lip rounding was perceived in the auditory signal and exploited to appropriately identify the upcoming rounded vowel. Further examination of the identification functions in figure 2 suggests that sighted participants have a steeper identification slope than blind participants. In order to extract identification boundaries Perception of Rounding in Blind Speakers

Blind Boundary

–12

Slope

–7 –2 i/y category boundary relative to /z/ offset (in 16-ms frames)

3

Fig. 3. Fifty percent category boundary (solid line) and slope (dashed line) of the identification function of the /izy/ sequence for blind and sighted participants. The category boundary is measured in terms of 16-ms-long frames. Error bars are standard errors.

and slopes for each participant, Probit analyses [14, 20] were performed on each speaker’s identification responses. Across speakers, the modeled labeling functions significantly predicted the actual data [for sighted participants, results ranged from χ2 (14, 42.77), p < 0.01, to χ2 (14, 83.78), p < 0.01; for blind participants, results ranged from χ2 (14, 50.23), p < 0.01, to χ2 (14, 96.02), p < 0.01]. Figure 3 depicts the average values of boundaries and slopes, for both speaker groups. Error bars are standard errors. Two one-way ANOVAs were conducted separately with identification boundaries and identification slopes as the dependent variables and group as the independent variable. Blind and sighted participants did not differ significantly in terms of identification boundaries, but sighted participants had significantly steeper identification slopes than blind participants [F(1, 32) = 11.25; p < 0.01; partial η2 = 0.35]. Acoustic Cues to Anticipatory Rounding Since the identification of the rounded vowel /y/ is possible (at a 50% threshold) during the production of the /z/ consonant, we investigated the nature of the acoustic anticipatory cues to rounding. As can be expected, and as Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

87

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

Results

Group

Sighted

displayed in figure 2, those acoustic measures highly covary [F(1, 10) = 113.387; p < 0.001]. It can be seen that the 50% boundary (dotted arrow in fig.  2) occurs at an F2 value of 3,028 Hz and a center of gravity value of 6,600 Hz. Compared to the consonant onset, those values correspond to a decrease of 450 Hz in the center of gravity and of 300 Hz in F2. These changes do not occur during the /izi/ sequences, and thus, do not involve the perception of the rounded vowel /y/.

Discussion

The present paper aimed to investigate the auditory perceptual ability to detect coarticulatory rounding cues by sighted and congenitally blind francophone adults. Participants had the task of identifying the upcoming vowel in various truncated versions of the /izy/ and /izi/ sequences. Results showed that blind and sighted speakers could identify the upcoming rounded vowel well before its acoustic onset. The 50% boundary of the identification functions did not differ significantly between groups, revealing that sighted speakers could perceive the acoustic cues associated with rounding as well as the blind speakers in the auditory modality, a result that agrees with previous data [14]. The category boundary corresponds to a 400-Hz decrease in the fricative center of gravity and a decrease of 300 Hz in F2 during the consonant. However, our results differ from those of Hirsch et al. [11] who suggest that blind listeners can identify the rounded vowel earlier than their sighted peers in truncated versions of the /isy/ sequence. However, it is not clear in the study of Hirsch et al. [11] whether individual listeners’ labeling functions

were computed, or if listeners’ responses were considered only on a group basis. Moreover, no statistical analysis was performed in the latter paper. Concerning the slope of the labeling functions, in our experiment, a significant effect of speaker group was found, with sighted speakers having steeper slopes (more categorical) than congenitally blind speakers. Thus, both speaker groups made use of anticipatory cues, but in slightly different ways. The difference in identification slopes likely suggests that sighted speakers display finer abilities to perceptually follow the decreasing values of the frication noise, compared to blind speakers, in this specific set of syllables. In blind speakers, the reduced ability to perceptually follow the falling movement of the frication noise may be the consequence, at the perceptual level, of a limited control in producing the rounding gesture, a visually and auditory cued gesture [10, 12, 13]. Also, this might suggest that, in relating F2 changes to their perceptual consequences, sighted speakers have acquired more categorical representations because they have associated some of these changes with lip rounding gestures that are visible. Blind speakers’ perception of F2 changes might be less categorical (when associated with lip gestures) likely because no visual component is associated with them. This interpretation is, however, speculative. Further studies are currently being conducted to explore the production-perception relationships in congenitally blind speakers. Acknowledgements The authors thank Marlene Busko for copy editing the paper.

References

88

4 Benguerel A-P, Adelman S: Perception of coarticulated lip rounding. Phonetica 1976; 33: 113–126. 5 Hecker V, Vaxelaire B, Cathiard M-A, Savariaux C, Sock R: How lip protrusion expansion influences auditory perceptual extent. Probing into the movement expansion model; in Cavé C, Guaïtella I, Santi S (eds): Oralité et gestualité: Interactions et comportements multimodaux dans la communication. Paris, L’Harmattan, 2001, pp 450–456. 6 Sock R, Hecker V, Cathiard M-A: The perceptual effects of anticipatory vocalic gestures in French. Proc XIVth Int Congr Phonet Sci, San Francisco, 1999, vol 3, pp 2057–2060.

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

7 Hirsch F, Sock R, Roy J-P, Canault M: La perception catégorielle anticipatoire des voyelles arrondies du français. Proc J Etude Parole, Fès, 2004. 8 Abry C, Boë L-J, Schwartz J-L: Plateaus, catastrophes and the structuring of vowels systems. J Phonet 1989;17:47–54. 9 Troille E, Cathiard M-A, Abry C: Speech face perception is locked to anticipation in speech production. Speech Commun 2010; 52: 513– 524. 10 Ménard L, Dupont S, Baum SR, Aubin J: Production and perception of French vowels by congenitally blind adults and sighted adults. J Acoust Soc Am 2009;126:1406–1414.

Ménard/Cathiard/Troille/Giroux

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

1 Blumstein S, Stevens K: Perceptual invariance and onset spectra for stop consonants in different vowel environments. J Acoust Soc Am 1980;67:648–662. 2 Ohde R, Sharf D: Order effect of acoustic segments of VC and CV syllables on stop and vowel identification. J Speech Hear Res 1977; 20:543–554. 3 Cathiard M-A, Lallouache M-T, Abry C: Does movement on the lips mean movement in the mind?; in Stork D, Hennecke M (eds): Speechreading by Humans and Machines. NATO ASI Series F: Computer and Systems Sciences. Springer, Berlin, 1996, vol 150, pp 211–219.

Perception of Rounding in Blind Speakers

14 Troille E, Cathiard M-A, Abry C, Ménard L, Beautemps D: Multimodal perception of anticipatory behavior. Comparing blind, hearing and cued speech subjects. Proc Audiovis Speech Processing Conf, Australia, 2008, pp 63–68. 15 Nittrouer S, Studdert-Kennedy M, McGowan R: The emergence of phonetic segments: evidence from the spectral structure of fricativevowel syllables spoken by children and adults. J Speech Hear Res 1989;32:120–132. 16 Munson B: Ariability in /s/ production in children and adults: evidence from dynamic measures of spectral means. J Speech Lang Hear Res 2004;47:58–69.

17 Nissen SL, Fox RA: Acoustic and spectral characteristics of young children’s fricative productions: a developmental perspective. J Acoust Soc Am 2005;118:2570–2578. 18 Wrench A: Analysis of fricatives using multiple centres of gravity. Proc Int Conf Phonet Sci 1995;4:460–463. 19 Forrest K, Weismer G, Milenkovic P, Dougall P: Statistical analysis of word-initial voiceless obstruents: Preliminary data. J Acoust Soc Am 1988;84:115–123. 20 Finney D-J: Probit Analysis, ed 3. Cambridge, Cambridge University Press, 1971.

Folia Phoniatr Logop 2015;67:83–89 DOI: 10.1159/000434719

89

Downloaded by: Siriraj Medical Library, Mahidol University 198.143.39.97 - 2/13/2016 7:45:09 AM

11 Hirsch F, Dreyfus H, Sock R, Vaxelaire B, Fauth C, Bouarourou F, Béchet M: Etude préliminaire de la perception précoce des voyelles labialisées par des auditeurs déficients visuels. Proc J Etude Parole, Mons, 2010, pp 137–140. 12 Ménard L, Leclerc A, Tiede M: Articulatory and acoustic correlates of contrastive focus in congenitally blind adults and sighted adults. J Speech Lang Hear Res 2014;57:793–804. 13 Ménard L, Toupin C, Baum S, Drouin S, Aubin J, Tiede M: Acoustic and articulatory analysis of French vowels produced by congenitally blind adults and sighted adults. J Acoust Soc Am 2013;134:2975–2987.

Effects of Congenital Visual Deprivation on the Auditory Perception of Anticipatory Labial Coarticulation.

It has been shown previously that congenitally blind francophone adults had higher auditory discrimination scores than sighted adults. It is unclear, ...
566B Sizes 0 Downloads 4 Views