Acoustic properties of fricatives /s/ and /∫/ produced by speakers with apraxia of speech: Preliminary findings from Arabic

1.1 Literature review

Fricative production in AOS is shaped by motor control deficits, influencing the turbulence and spectral properties of airflow during articulation. The spectral characteristics of fricatives are largely determined by the place and manner of constriction within the oral cavity (Haley et al. 2023, Jongman et al. 2000). Sibilant fricatives, particularly /s/ and /ʃ/, generate high-intensity aperiodic energy, with their acoustic profiles shaped by articulatory configuration and movement (Bunker et al. 2024, Haley et al. 2010).

A key characteristic of apraxic speech is a reduction in fricative intensity, weakening phonetic contrast and speech clarity (Allison et al. 2020, Duffy et al. 2020). Specifically, the intensity contrast between fricatives and following vowels, typically ranging from 7 to 14 dB, serves as a crucial perceptual cue for speech intelligibility (Kim 2017). Additionally, spectral mean or COG has been identified as a reliable acoustic measure for analyzing fricative production, surpassing traditional metrics like peak frequency in sensitivity to articulatory impairments (Forrest et al. 1988, Jongman et al. 2000, Park 2017).

Spectral moments, particularly COG, provide critical insights into fricative articulation by capturing constriction length and shape. Generally, higher COG values indicate a more anterior constriction – as observed in /s/, while lower values correspond to more posterior constrictions, as in /ʃ/ (Nittrouer 1995, Park 2017). The postalveolar fricative /ʃ/ exhibits a lower spectral mean than the alveolar /s/ due to a more posterior tongue placement and frequent lip rounding (Maas and Mailend 2017). Research further indicates that voiceless fricatives exhibit greater energy concentration in higher frequencies than their voiced counterparts (Nittrouer 1995).

Studies on vowel production in AOS report inconsistent findings, with some showing normal formant values, while others indicate reduced vowel space and centralization, particularly in multisyllabic contexts (Kent and Rosenbek 1983, Ryalls 1986, Haley et al. 2001). Haley et al. (2001) observed that speakers with AOS either maintained normal vowel formant ranges or exhibited a contracted vowel space, suggesting variability in articulatory precision.

The relationship between spectral peak location and place of articulation further highlights articulatory instability in AOS (Shadle 2012). For instance, the spectral peak of /s/ typically falls between 3.5 and 5 kHz, whereas /ʃ/ ranges from 2.5 to 3.5 kHz. In addition, sibilant fricatives (/s, z/) generally have higher amplitude than nonsibilants (/f, v/), reflecting their stronger energy concentration (Shadle 2012).

In individuals with AOS, fricative production is often imprecise, leading to reduced acoustic differentiation and variability in /s/ and /ʃ/ contrasts (Duffy 2006). Repetitive productions and articulatory overlap have been reported, further illustrating inconsistent spatial targeting during speech (Haley et al. 2001). Ryalls (1986) noted that speakers with AOS exhibited higher formant values than control participants, likely reflecting difficulties in achieving smooth consonant-vowel transitions, reinforcing the motor coordination deficits characteristic of the disorder (Pickett 1999).

By using spectrographic and linear predictive coding analyses, Harmes et al. (1984) examined temporal and articulatory control in fricative production among individuals with AOS. Their findings revealed significant spectral shifts and altered energy concentration frequencies, underscoring the distinct acoustic patterns associated with motor impairments.

3.1 Participants

This preliminary study reports data from two Palestinian Arabic-speaking male individuals with AOS, referred to here as patients B1 and C1. Patient B1 is 71 years old, and patient C1 is 66 years old; both are right handed and college educated. Both participants experienced an infarction in the left middle cerebral artery, with 68 and 75 months elapsed since the cerebrovascular accident for patients B1 and C1, respectively, as presented in Table 1.

The small sample size poses limitations on the generalizability of the findings, yet this study offers crucial preliminary insights into AOS in an underrepresented population. Expanding participant recruitment in Palestine presents distinct challenges, including restricted institutional support for accessing this population and cultural sensitivities surrounding research participation. Researchers frequently depend on personal networks to identify potential participants, as individuals with AOS and their families may be reluctant to engage in such studies. Furthermore, the relatively low prevalence of AOS further constrains participant availability, complicating efforts to assemble a broader sample.

Participants were selected based on the AOS diagnostic criteria established by Staiger et al. (2012), which required (a) a confirmed diagnosis of AOS, (b) absence of dysarthria, (c) minimal or no aphasic impairment, (d) normal hearing and perceptual abilities, and (e) the ability to repeat multisyllabic words and phrases without examiner cueing.

AOS diagnosis was confirmed based on the presence of key clinical features observed during assessment tasks, including (a) slow speech with initiation difficulties, (b) self-corrections and audible groping, (c) segment lengthening and syllable segregation, (d) predominance of sound distortions, (e) prosodic impairments, and (f) a mixed phonetic-phonemic deficit profile (Wambaugh et al., 2013). Both patients fully met the inclusion and diagnostic criteria. Hearing was normal, and no signs of dysarthria or visual impairments were detected.

To ensure diagnostic accuracy, the Apraxia Battery for Adults – Second Edition (ABA-2) (Dabul 2000) was used to assess motor speech deficits, including diadochokinetic rates, increasing word length effects, and repeated trials. The diagnosis was further supported by a perceptual speech evaluation, conducted by two experienced speech-language pathologists (SLPs) with expertise in motor speech disorders. These SLPs independently assessed the participants’ speech and later reached a consensus to confirm the diagnosis.

To exclude dysarthria, a structural and functional examination was conducted, assessing (a) the oral mechanism (jaw, lips, and tongue), (b) velopharyngeal function, (c) respiration, and (d) phonation (Bislick and Hula 2019). In addition, Broca’s aphasia was ruled out based on performance on the standardized Arabic version of the Western Aphasia Battery (WAB-A) (Manaf and Shetty 2022).

Speech intelligibility was analyzed using broad transcription by two external SLPs, who conducted a perceptual judgment of speech during the recording session. Following Jacks (2006), target words assessed the intelligibility of individuals with AOS, as detailed in Table 2. Speech intelligibility was defined as the degree to which listeners accurately perceived the speaker’s intended words (Kim and Kuo 2011). Errors were identified by comparing the intended utterances with the perceived consonant types of errors.

Speech intelligibility was measured as the percentage of target words correctly transcribed by the SLPs. The phonetic match percentage was calculated based on the number of correctly identified phonemes relative to the total number of phonemes in the target words, as presented in Table 2. The SLPs also used a three-point rating scale to evaluate impairment severity (1 = mild, 2 = moderate, 3 = severe), with both participants rated as moderate in overall severity. In addition, two healthy Palestinian Arabic speakers, matched to the AOS participants in gender, age, and dialect, served as the control group.

3.2 Speech sample

Table 3 presents the stimuli used in the study, while Figures 1 and 2 display production samples from the speaker groups. For the repetition task, the speech sample consisted of the fricatives /s/ and /ʃ/ followed by long vowels (/aː/, /uː/, /iː/) embedded in consonant–vowel–consonant words. Repetition was chosen to minimize potential comprehension and reading difficulties that individuals with AOS may face (Kempler and Van Lancker 2002). Real words were used whenever possible; when unavailable, pseudowords with the same structure were substituted. To elicit the target words, a carrier phrase, ‘ʔiħki’ (‘Say’), was used. The stimuli were printed on 20 cm × 20 cm cards and shown to participants.

Figure 1

A sample of the production of the syllable /∫u/ by an individual with AOS.

Figure 2

A sample of the production of the syllable /∫u/ as produced by a control speaker.

A Palestinian Arabic-speaking SLP pronounced each target word, which participants repeated five times, resulting in 120 utterances per group (12 words × 5 repetitions × 2 participants). Healthy speakers produced the same number of repetitions. To ensure consistency, participants were instructed to maintain a typical pitch and loudness. A brief practice session was conducted before recording to familiarize participants with the task.

3.3 Recording procedures

Recordings were conducted in a quiet room with a high-quality AKG Perception 120 USB microphone, placed 15 cm from the participants’ mouths to minimize distortion. Speech samples were captured at 44 kHz (16-bit quantization) and low-pass filtered at 16 kHz for acoustic analysis. If a participant had difficulty producing a target word, the examiner would repeat the prompt before resuming the recording.

3.4 Acoustic measurements

Duration measurements for the initial fricatives and vowel segments were conducted using a wideband spectrogram and waveform. The onset and offset of noise were used as reference points to mark the beginning and end of each fricative. The fricative onset was identified by the concentration of high-frequency energy in the spectrogram, while the fricative offset was defined as the point of minimum intensity immediately before the onset of the following vowel (Jongman et al. 2000). Thus, fricative duration was measured as the interval between the fricative’s onset and offset.

Following Haley et al. (2001), mean spectral analysis was conducted using a 20-ms window centered 50 ms after fricative onset, employing PRAAT software (Boersma and Weenink 2020). This interval, shown to provide reliable spectral distinctions for AOS-related studies, was selected to enhance the differentiation of fricative sounds. Formant frequencies (F1 and F2) for vowels following the initial fricatives were automatically extracted using PRAAT and Phono Lab software (Metoui 1995).

To plot the acoustic vowel space, formant normalization was performed using the Visible Vowels software (Heeringa and van de Velde 2020) to reduce inter-speaker variability due to physiological differences in vocal tract configuration (Vorperian and Kent 2014). The acoustic analysis applied Lobanov’s z-score normalization method (Lobanov 1971). The COG, or first spectral moment (M1), in Hz, was calculated using TF32 software, based on three 50-ms windows of fricative noise with a 25-ms overlap between successive windows (Kim 2017).

The use of three overlapping 50-ms windows improved temporal resolution, enabling a more precise capture of spectral changes within the fricative sounds. The fricative intensity was measured by calculating the average intensity of each test segment in decibels (dB), and the intensity difference between the fricative and the following vowel was used as a dependent variable.

3.5 Statistical analysis

A two-way analysis of variance (ANOVA) was conducted to examine the effects of speaker group (AOS vs control) and fricative type (/s/ vs /ʃ/) on segment duration, intensity, spectral characteristics, and formant frequencies (F1, F2, and F3). For vowel-related measures, vowel category (/aː/, /uː/, /iː/) was included as an additional independent factor. The analysis assessed main effects and interactions for each acoustic measure. Post hoc tests were not conducted because the ANOVA did not reveal significant interactions that warranted further pairwise comparisons. Instead, we interpreted the main effects directly based on the ANOVA results.

4.1 Segment duration

Table 4 presents the mean durations (in milliseconds) of fricatives and vowels, along with their corresponding intensity values (dB). A two-way ANOVA revealed a significant main effect of the speaker group (F(1,10) = 123.884, p < 0.001), with AOS participants producing significantly longer fricative durations compared to controls. A main effect of fricative type was also found (F(1,10) = 70.098, p < 0.05), indicating that /ʃ/ was consistently longer than /s/ in both groups. However, no significant interaction effect was observed between speaker group and fricative type (F(1,10) = 77.233, p > 0.05), suggesting that AOS participants maintained durational contrasts between the two fricatives.

Vowel duration also differed significantly between groups. Levene’s test for equality of variances showed significant differences following /s/ (F(1,10) = 38.423, p < 0.001) and /ʃ/ (F(1,10) = 0.043, p < 0.001), with AOS speakers producing longer vowels than controls. A two-factor ANOVA confirmed a significant effect of place of articulation on fricative duration for both /s/ (F(1,10) = 7.706, p = 0.02) and /ʃ/ (F(1,10) = 302.001, p < 0.001), reinforcing the idea that motor planning impairments in AOS contribute to prolonged segment durations.

4.2 Intensity

The intensity of vowels and fricatives was analyzed and compared between healthy speakers and speakers with AOS, as shown in Table 4 and Figures 3 and 4. Levene’s test for equality of variances revealed a significant difference between the speaker groups in terms of vowel intensity following the fricatives /s/ (F(1,10) = 0.454, p < 0.001) and /ʃ/ (F(1,10) = 0.887, p < 0.001). Analysis of fricative intensity also showed significant differences between healthy speakers and speakers with AOS for both /s/ (F(1,10) = 0.092, p < 0.001) and /ʃ/ (F(1,10) = 1.269, p < 0.001).

Figure 3

Intensity values (dB) for /s/ and /ʃ/ as produced by control speakers and speakers with AOS.

Figure 4

Vowel intensity values (dB) following /s/ and /ʃ/ as produced by control speakers and speakers with AOS.

For speakers with AOS, the intensity difference was more pronounced for /s/, whereas for healthy speakers, it was greater for /ʃ/. A two-factor ANOVA revealed a significant effect on vowel intensity following /s/ (F(1,10) = 61.936, p < 0.001) and /ʃ/ (F(1,10) = 74.711, p < 0.001). Similarly, significant effects were found for the intensity of the fricatives themselves, with both /s/ (F(1,10) = 346.782, p < 0.001) and /ʃ/ (F(1,10) = 63.298, p < 0.001) showing significant results.

A two-way ANOVA revealed a significant effect of fricative type (F(1,10) = 44.356, p < 0.05), indicating that mean spectral peaks differed between /s/ and /ʃ/ across speaker groups. In addition, individuals with AOS exhibited significantly lower mean spectral peaks compared to controls (F(1,10) = 5.087, p < 0.01). Further analysis confirmed group differences for both /s/ (F(1,10) = 77.22, p < 0.001) and /ʃ/ (F(1,10) = 40.798, p < 0.001), suggesting reduced articulatory precision in AOS.

Despite lower spectral peaks, AOS speakers maintained similar overall spectral patterns to the control group. No significant interaction effect of fricative type and speaker group was found (F(4.41) = 55.10, p > 0.05), indicating consistent spectral differences across conditions. As expected, the spectral mean of /s/ was consistently higher than /ʃ/ in both groups. Specifically, AOS speakers exhibited acoustic energy concentrations for /s/ in the 1,100–4,900 Hz range, while for /ʃ/, energy ranged from 830 to 4,900 Hz. In contrast, healthy speakers showed higher energy distributions, with /s/ ranging from 5,000 to 6,000 Hz and /ʃ/ between 4,800 and 5,980 Hz. These findings align with previous research indicating that sibilant fricatives exhibit robust spectral and temporal properties (Harmes et al. 1984), particularly in healthy speech, where /s/ demonstrates a strong high-frequency concentration.

The formant frequencies (F1, F2, and F3) were analyzed for all vowels and compared between AOS and control speakers. Only significant and relevant differences are reported. Figures 5–7 illustrate the F1, F2, and F3 values for each vowel category across groups.

Figure 5

F1 values (Hz) for the vowels (/a:, u:, i:/) following /s/ and /∫/ produced by the control group and speakers with AOS.

Figure 6

F2 values (Hz) for the vowels (/a:, u:, i:/) following /s/ and /∫/ produced by the control group and speakers with AOS.

Figure 7

F3 values (Hz) for the vowels (/a:, u:, i:/) following /s/ and /∫/ produced by the control group and speakers with AOS.

For /aː/, Levene’s test for F1 variances showed no significant difference between groups (F(1,10) = 0.015, p > 0.05). However, a two-way ANOVA revealed a significant main effect of the speaker group on F1 (F(1,10) = 411.897, p < 0.001), indicating altered tongue height in AOS. A similar trend was observed for F2, where Levene’s test suggested equal variances across groups (F(1,10) = 1.803, p > 0.05), but a significant between-group difference was detected (F(1,10) = 129.112, p < 0.001), suggesting changes in tongue advancement.

For /uː/, both F1 (F(1,10) = 33.015, p < 0.05) and F2 (F(1,10) = 18.642, p < 0.001) were significantly different between groups, reflecting potential instability in vowel articulation in AOS. Similarly, for /iː/, a significant effect was observed for F1 (F(1,10) = 54.399, p < 0.001), while F2 differences were nonsignificant (F(1,10) = 0.007, p > 0.05), indicating that tongue height was more affected than tongue advancement in AOS speakers.

F3 analysis revealed some differences between groups, but these were less consistent than F1 and F2. For /aː/, a two-way ANOVA showed a significant difference in F3 values (F(1,10) = 87.542, p < 0.001), suggesting variability in vocal tract shaping. However, for /iː/, the difference in F3 was only marginally significant (F(1,10) = 34.221, p < 0.01), and for /uː/, while differences were observed (F(1,10) = 42.378, p < 0.01), the variability in F3 across participants was higher than in F1 and F2. These results indicate that while AOS may influence finer articulatory adjustments reflected in F3, its role is less systematic compared to F1 and F2.

These findings suggest that individuals with AOS may have difficulty achieving accurate mouth positioning, which could lead to higher F1 values for the back vowel /uː/ compared to the front vowel /iː/ (p < 0.05). The significant differences in F3 indicate potential instability in vocal tract shaping, but the inconsistencies across vowels suggest that F3 is not as reliable as F1 and F2 in characterizing vowel production deficits in AOS. Despite these variations, speakers with AOS were able to produce distinct vowel categories without acoustic overlap, as shown in Figure 5.

4.3 COG

The analysis showed that the first spectral moment (M1) was higher for /s/ than for /ʃ/ in both speaker groups, indicating a main effect of the word. A two-way ANOVA revealed a significant effect for the COG and fricative type (F(4.567, 357.324) = 6.986, p > 0.05), suggesting that the mean COG values significantly differed between fricative types, regardless of the speaker group. An independent samples t-test further confirmed significant differences in COG mean values between the speaker groups, with /s/ (F(1,10) = 7.732, p < 0.05, P = 0.022) and /ʃ/ (F(1,10) = 13.311, p < 0.05, P = 0.004) both showing significant results. Compared to healthy speakers, individuals with AOS demonstrated lower intensity and lower COG mean values for both fricatives /s/ and /ʃ/, as shown in Figure 8.

Figure 8

Mean of fricative central gravity as produced by speakers with AOS and HS.