Atypical vocal imitation of speech and song in autism spectrum disorder: Evidence from Mandarin speakers

Participants

A group of 33 autistic children (aged between 7 and 16) and 30 age-matched non-autistic children took part in the study. All were native speakers of Mandarin and reported no history of other neurological or psychiatric disorders. They were recruited from special educational facilities and mainstream schools in Nanchang and Nanjing, China. The autistic children all had a clinical diagnosis of autism using either DSM-IV or DSM-5 (American Psychiatric Association, 1994, 2013) which was further supported by the Autism Diagnostic Observation Schedule—Second Edition (ADOS-2) (Lord et al., 2012) conducted by the first author (with research reliability for administration and scoring). All autistic participants were administered the ADOS-2 Module 3 according to their developmental and language levels. Total scores on the ADOS-2 were converted to a comparative score (CS) of 1–10, with 10 representing the highest severity of autism-related symptoms (Duda et al., 2014; Gotham et al., 2009). All participants had normal hearing in both ears, with pure-tone air conduction thresholds of 25 dB HL or better at frequencies of 0.5, 1, 2, and 4 kHz, as assessed using an Amplivox manual audiometer (Model 116). Participants completed a nonverbal IQ test using the Raven’s Standard Progressive Matrices Test (RSPM) (Raven et al., 1998) and a receptive vocabulary test using the Chinese version of the Peabody Picture Vocabulary Test-Revised (PPVT-R) (Dunn & Dunn, 1981; Sang & Miao, 1990). The standardized scores for RSPM and PPVT-R were calculated as described by Wang et al. (2023). For RSPM, the standardized scores were derived using the means and standard deviations from a Chinese normative study (Zhang, 1989). As the Chinese norms for PPVT-R covered only ages 3.5 to 9 (Sang & Miao, 1990), we used American norms (Dunn & Dunn, 1981) to calculate the standardized scores. A correlation analysis showed a significant positive relationship (r = 0.95) between the standardized scores based on the Chinese norms and those based on the American norms for participants aged 9 and below, validating this methodology. The Chinese version of the forward digit span task was used to assess verbal short-term memory (Wechsler, 2003). Participants’ musical training background and their ability to identify a musical note without a reference tone (i.e. absolute pitch or perfect pitch) (Deutsch, 2013) were collected using a caregiver-reported questionnaire, and their years of formal musical training were summed across all instruments including voice (Wang, Beaman, et al., 2021). Participants’ perceptual skills were assessed using a statement-question intonation discrimination task, taken from a comparative study investigating speech and music perception (F. Liu et al., 2012; Wang et al., 2023). As can be seen in Table 1, the results of Welch’s t-test showed that the autistic and non-autistic groups were comparable on all background measures, except the PPVT-R scores, which were taken into account in the statistical models.

OPEN IN VIEWER

Table 1.

Characteristics of the autism (n = 33) and non-autism groups (n = 30).

Background measures	Autism	Non-Autism	t	p	Cohen’s d
Gender (F:M)	5:28	4:26
Age	10.29 (2.50)	11.50 (2.83)	1.79	0.08	0.45
Musical training	0.88 (1.32)	0.50 (1.11)	1.24	0.22	0.31
RSPM	110.12 (15.77)	112.72 (10.26)	0.78	0.44	0.20
PPVT-R	124.33 (25.87)	141.77 (12.80)	3.44	0.001**	0.85
Digit span	8.49 (0.91)	8.07 (1.11)	1.63	0.11	0.41
Self-reported absolute pitch	n = 2	n = 3
Perception-Natural speech	1.57 (0.87)	1.81 (0.75)	1.19	0.24	0.30
Perception-Gliding tone	1.60 (0.86)	1.93 (0.63)	1.71	0.09	0.43
ADOS-CS	6.97 (2.31)	NA

Note. Musical training: years of musical training; RSPM: standard score of Raven’s Standard Progressive Matrices Test; PPVT-R: standard score of Peabody Picture Vocabulary Test-Revised; Digit span: raw score of verbal short-term memory; Perception-Natural speech and Perception-Gliding tone: D-prime values for subtest scores, with higher values representing better perception skill; ADOS-CS: comparative score of ADOS, with 10 representing the highest severity of autism-related symptoms. Bold values indicate statistical significance at p < 0.05. *p < 0.05, **p < 0.01, ***p < 0.001.

Community involvement

There was no community involvement in the present study.

Stimuli

The model stimuli were 10 sentences either spoken or sung with an early focus or a late focus from F. Liu et al. (2013), yielding 40 sentences with two to six syllables each (see Table 2 for the list of sentences and Supplementary Table 1 for musical notations of the sung stimuli). The inclusion of different sentence lengths was to control for the effect of stimulus length on imitation performance (F. Liu et al., 2013). The manipulation of the different focus conditions of the sentences ensured the inclusion of a variety of pitch and duration patterns in the speech stimuli, as focused words normally show a higher pitch and longer duration than their unfocused counterparts in Mandarin speech (F. Liu & Xu, 2005; Yuan, 2011). For example, in the top right panel of Figure 2, the sentence “她的包?” [“Her bag?”] has an initial focus on the word “她” [“Her”] which has a higher pitch and longer duration than the same unfocused word in the bottom right panel of Figure 2, where the sentence “她的包?” [“‘Her bag?”] has a final focus on the word “包” [“bag”]. As can be seen from the top and bottom left panels of Figure 2, the corresponding song stimuli approximated the global melodic contours and timing variations of the speech stimuli. To both accommodate participants’ vocal range and to ensure that participants of different ages or gender were exposed to the same pitch and duration patterns of the speech/song stimuli, we adopted the male and female versions of the stimuli from F. Liu et al. (2013). The female model was originally recorded by a 27-year-old Mandarin-speaking female student who was born and raised in Beijing. To ensure that the stimuli encountered by male and female participants have identical pitch intervals and rhythmic patterns, the female model was synthesized (preserving the absolute pitches and formant frequencies of the original recordings) and the male model was generated from the female model by changing the original pitches to one octave lower and shifting the frequencies of the original formants by .78 to achieve male voice characteristics, using the “change gender” command in Praat (Boersma & Weenink, 2001). The ecological validity of the synthesized female and male models was tested and confirmed in F. Liu et al. (2013), where Mandarin-speaking female and male adult participants with and without congenital amusia performed the same imitation task using the same stimulus set. None of the participants in F. Liu et al. (2013) noted any unnaturalness of the stimuli, and no significant differences were found in imitation performance between the participants of different genders for either the amusic or the non-amusic group. Thus, the current study adopted the same stimulus set as in F. Liu et al. (2013). We also did not observe any significant differences in imitation performance across female and male participants in the current sample (see Supplementary Table 2). The male version was used for male participants ⩾12 years old, and the female version was used for female participants regardless of age as well as male participants < 12 years old, as research indicates that children up to 12 show similar pitch ranges (Mecke & Sundberg, 2010; Nicollas et al., 2008; Sergeant & Welch, 2009).

OPEN IN VIEWER

Table 2.

Stimuli used in the experiment.

Stimuli with an early focus	Stimuli with a late focus	Chinese Pinyin	English translation
黑车？	黑车？	Hei1 che1?	Black car?
青天？	青天？	Qing1 tian1?	Blue sky?
她的包？	她的包？	Ta1 de0 bao1?	Her bag?
三颗星？	三颗星？	San1 ke1 xing1?	Three stars?
冬天的风？	冬天的风？	Dong1 tian1 de0 feng1?	The winter’s wind?
写他书上？	写他书上？	Xie3 ta1 shu1 shang0?	Write on his book?
漆黑的天空？	漆黑的天空？	Qi1 hei1 de0 tian1 kong1?	Pitch-black sky?
小丁长高了？	小丁长高了？	Xiao3 ding1 zhang3 gao1 le0	Xiao Ding grew taller?
老郭的猫丢了？	老郭的猫丢了？	Lao3 guo1 de0 mao1 diu1 le0	Lao Guo’s cat is lost?
小方天天加班？	小方天天加班？	Xiao3 Fang1 tian1 tian1 jia1 ban1?	Xiao Fang works overtime every day?

OPEN IN VIEWER

Figure 2.

The pitch-time trajectory of the sentence “她的包? vs. 她的包? (Ta1 de0 bao1?/ Her bag?) under different conditions by female/male model speakers.

Procedure

The presentation of the model stimuli and the recording of the imitations were both done using Praat (Boersma & Weenink, 2001). Participants were seated in a quiet room and were presented with four practice trials (with items different from those in experimental trials: 2 speech vs 2 song) to familiarize themselves with the task and the recording environment. Following the practice section, participants were presented with each of the 40 speech/song sentences one at a time in pseudorandom order and were instructed to imitate exactly the pitch and timing patterns of the sentences to the best of their ability, while their voices were recorded via a Roland RUBIX22 USB Audio Interface. Each sentence was played once and only replayed when participants failed to catch the words, and not when they wanted to listen to it again so they could imitate it better.

Data analysis

Recordings were analyzed in Praat using ProsodyPro, a software tool designed for the automatic analysis of extensive speech data (Xu, 2013). To ensure precise acoustic measurements, we adopted a hybrid approach. This involved initial automated processes using ProsodyPro and subsequent manual verification by trained phoneticians (authors LW and FL) to extract the pitch and duration of each syllable rhyme. Syllable/note duration was calculated as the length of the syllable rhyme, and the onset of syllable rhyme was defined as the syllable/note onset time. The median F0s (fundamental frequencies) of the syllable rhymes were extracted to indicate pitch heights. Octave errors in pitch imitation were corrected, that is, when the imitated pitch was more than 6 semitones (half octave) apart from the model pitch, the value was adjusted as 12—imitated pitch. In total, less than 4.11% of the data samples needed to be adjusted, and most of these errors were caused by creaky voices, resulting in decreased F0 (Johnson, 2011). Trained phoneticians manually added these missed vocal pulse marks for F0 based on the waveforms and spectrograms, to avoid having erroneous outliers misleading imitation results.

We used absolute pitch and duration matching to refer to the ability to imitate individual syllables/notes based on their acoustic features, irrespective of their relationship with surrounding syllables/notes. In addition, following F. Liu et al. (2013) and previous singing or pitch-matching studies (Dalla Bella et al., 2007, 2009, 2011; Pfordresher & Brown, 2007; Pfordresher et al., 2010; Ward & Burns, 1978), we also measured the number of pitch contour, pitch interval, and time errors that deviated from the corresponding model’s pitch direction or specific pitch interval or duration value. The pitch was measured in “cents” (100 cents = one semitone), a unit of measure based on the equal-tempered scale in music, to facilitate a more nuanced representation of pitch distinctions and a finer resolution in the assessment of pitch differences. Detailed definitions of these measures are provided below.

The absolute pitch deviation (in cents): Median F0 was extracted from each syllable rhyme and then subtracted from that of their matched model to find the pitch deviation (in absolute value) for each imitated rhyme. The deviations were averaged over all syllables/notes in each utterance/melody and the bigger the value, the less accurate the imitation in terms of absolute pitch matching.

The relative pitch deviation (in cents): The pitch interval was calculated as the absolute difference in median F0 between two consecutive syllables/notes, and then subtracted from their matched model’s pitch interval (in absolute value). The deviations were averaged over all intervals in each utterance/melody and the bigger the value, the less accurate the imitation in terms of relative pitch matching.

The number of pitch contour errors: Pitch contour errors were defined as imitated pitch intervals that differed from the corresponding model pitch intervals regarding pitch directions (up, down, or level). Pitch direction was considered to be up or down if the difference in pitch interval was higher or lower by 50 cents or more; otherwise (the difference was within 50 cents), the pitch intervals were considered to form a level/flat pitch direction. The number of contour errors was summed over each utterance/melody.

The number of pitch interval errors: Pitch interval errors were defined as imitated pitch intervals that were larger or smaller than the corresponding model pitch intervals by 100 cents without considering the pitch direction. Specifically, imitated and model pitch intervals were compared using absolute values. The number of pitch interval errors was summed over each utterance/melody.

The absolute duration difference (in milliseconds): Duration was extracted from each syllable rhyme and then subtracted from their matched model’s production to find the absolute difference for each rhyme. The differences were averaged over all rhymes in each utterance/melody and the larger the value, the less accurate the imitation in terms of absolute duration matching.

The relative duration difference (in milliseconds): Interonset interval (IOI) was calculated as the difference between the onsets of two consecutive syllables/notes, and then subtracted from their matched model’s IOI (in absolute value). The differences were averaged over all IOIs in each utterance/melody and the larger the value, the less accurate the imitation in terms of relative duration matching.

The number of time errors: Time errors were defined as imitated syllables/notes that were more than 25% longer or shorter than the corresponding model syllables/notes (Dalla Bella et al., 2007, 2009; Prince & Pfordresher, 2012). In Western tonal music, the durations of different events such as sixteenth notes (1/4 a beat), eighth notes (1/2 a beat), and quarter notes (1 beat) are in simple integer ratio relationships (Drake & Palmer, 2000). Similarly, speech rhythm can also be measured in relative terms (Patel & Daniele, 2003; Patel et al., 2006). Thus, using a 25% deviation to count time errors not only captures the violation of the time signature in music but also makes the comparison of spoken and musical rhythm possible. The number of time errors was summed over each utterance/melody.

All statistical analyses were conducted using Rstudio (RStudio Team, 2020). We performed linear mixed-effects analysis, which is robust to violations of statistical assumptions (Gelman & Hill, 2006; Schielzeth et al., 2020). The lme4 (Bates et al., 2012; Brauer & Curtin, 2018) and lmerTest (Kuznetsova et al., 2017) packages were used with the above-mentioned pitch and duration variables as the dependent variable and Group (effect-coded: Non-autism vs Autism), and Condition (effect-coded: Speech vs Music) as well as the interaction between Group and Condition as fixed effects. To take into account the significant group differences in receptive vocabulary and the relatively wide age range, we also added PPVT-R scores and age (both variables were mean-centered) in the models. Years of musical training were significantly associated with only one of the pitch metrics: More musical training was associated with fewer pitch interval errors (B = -0.06, SEB = 0.03, t(61.41) = -2.26, p = 0.03). Thus, in the interest of space, musical experience was not considered in the models. All models were fit using the maximal random effects structure that converged with two random factors (subject vs item) (Barr, 2013; Barr et al., 2013). When the maximal model failed to converge, the random correlations were removed first. If the model still failed to converge, the random effect with the lowest variance was iteratively removed until the model converged. Subsequent post hoc comparisons, if any, were conducted using the emmeans package with Holm-Bonferroni correction for multiple comparisons (Lenth et al., 2018).

Absolute pitch deviation

Figure 3(a) shows the distribution of absolute pitch deviations for each group in both the Speech and the Music conditions. These values were obtained by averaging the absolute pitch deviations across the syllables/notes (ranging from two to six) within each of the 20 utterances/melodies produced by each participant. These averages captured participants’ performance across the entire stimuli while minimizing the variations caused by extreme values (e.g. due to creaky voice). Results revealed a main effect of Condition (B = −22.55, SEB = 5.49, t(31.54) = −4.11, p < 0.001) and a Group * Condition interaction (B = −7.78, SEB = 3.36, t(51.99) = −2.31, p = 0.02). Post hoc analyses with Holm-Bonferroni correction for multiple comparisons suggested no group differences in either condition (Speech: t(72.8) = 0.17, p = 0.88; Music: t(72.6) = 1.67, p = 0.10); instead, the interaction was driven by both groups performing better on absolute pitch matching for music than for speech, with the trend being more pronounced in the autism group (t(48.5) = 4.76, p < 0.001, Music: M(SD) = 142.8(108.29); Speech: M(SD) = 201.96(110.98)) than in the non-autism group (t(50.9) = 2.27, p = 0.03, Music: M(SD) = 163.31(113.68); Speech: M(SD) = 192.22(112.98)). No other remaining main effects were significant (Table 3).

OPEN IN VIEWER

Figure 3.

Pitch-related measures for the autism and non-autism groups. (a) Absolute pitch deviations (in cents), with black lines representing mean values. (b) Relative pitch deviations (in cents), with black lines representing mean values. (c) The number of pitch contour errors, with error bars representing the standard deviation. (d) Number of pitch interval errors, with error bars representing the standard deviation. Different plots are selected depending on the nature of the data type, with (a) and (b) representing continuous data, (c) and (d) representing discrete data.

OPEN IN VIEWER

Table 3.

Coefficients for the linear mixed-effects models for pitch-related measures.

Measure	Effect	Estimate	Std. Error	df	t	p
Absolute pitch deviation	Group	–9.47	9.76	59.03	–0.97	0.34
	Condition	–22.55	5.49	31.54	–4.11	< 0.001***
	PPVT-R	–0.64	0.43	59.01	–1.48	0.14
	Age	–3.44	3.34	59.04	–1.03	0.31
	Group × Condition	–7.78	3.36	51.99	–2.31	0.02*
Relative pitch deviation	Group	7.04	5.25	58.70	1.34	0.19
	Condition	–44.77	9.02	23.89	–4.97	< 0.001***
	PPVT-R	–0.36	0.23	58.64	–1.55	0.13
	Age	–2.89	1.79	58.76	–1.61	0.11
	Group × Condition	–7.20	3.41	60.62	–2.11	0.04*
Pitch contour errors	Group	0.06	0.02	57.72	2.63	0.01*
	Condition	–0.18	0.04	24.82	–4.13	< 0.001***
	PPVT-R	–0.0006	0.001	57.64	–0.60	0.55
	Age	–0.009	0.008	57.86	–1.09	0.28
	Group × Condition	0.02	0.02	60.66	1.05	0.30
Pitch interval errors	Group	–0.01	0.04	55.82	–0.30	0.76
	Condition	–0.24	0.06	23.50	–4.16	< 0.001***
	PPVT-R	–0.002	0.002	58.91	–1.06	0.29
	Age	–0.03	0.01	59.10	–2.60	0.01*
	Group × Condition	–0.04	0.02	60.91	–1.68	0.10

Bold values indicate statistical significance at p < 0.05. *p < 0.05, **p < 0.01, ***p < 0.001.

Relative pitch deviation

Figure 3(b) shows the distribution of the relative pitch deviations for each group in both the Speech and the Music conditions. Results revealed a significant main effect of Condition (B = −44.77, SEB = 9.02, t(23.89) = −4.97, p < 0.001) and a significant interaction between Group and Condition (B = −7.20, SEB = 3.41, t(60.62) = −2.11, p = 0.04). Post hoc analyses with Holm-Bonferroni correction for multiple comparisons suggested that both groups showed better relative pitch matching for music than for speech (Autism: t(29.9) = 5.42, p < 0.001; Non-autism: t(31.1) = 3.88, p < 0.001), and the autism group performed worse than the non-autism group in the speech condition (t(102) = −2.27, p = 0.03, Autism: M(SD) = 215.08(133.52); Non-autism: M(SD) = 179.73(107.35)) but not in the music condition (t(102) = 0.03, p = 0.98, Autism: M(SD) = 113.88(86.71); Non-autism: M(SD) = 104.87(68.31)). No other remaining main effects were significant (see Table 3).

Number of pitch contour errors

Figure 3(c) shows the distribution of the number of pitch contour errors for each group in both the Speech and Music conditions. These values were obtained by summing errors over two to six syllables/notes within each of the 20 utterances/melodies produced by each participant. Results revealed, as shown in Table 3, significant main effects of Group (B = 0.06, SEB = 0.02, t(57.72) = 2.63, p = 0.01) and Condition (B = −0.18, SEB = 0.04, t(24.82) = -4.13, p < 0.001), as both groups made fewer contour errors with the music condition (Autism: M(SD) = 6.88(5.37), Non-autism: M(SD) = 3.20(2.73)) than the speech condition (Autism: M(SD) = 12.88(4.85), Non-autism: M(SD) = 11.33(3.90)), and the autism group exhibited more pitch contour errors than the non-autism group across both conditions. The interaction between Group * Condition and the effects of PPVT-R and Age were not significant.

Number of pitch interval errors

Figure 3(d) shows the distribution of the number of pitch interval errors for each group in both the Speech and Music conditions. As shown in Table 3, the linear mixed-effects model revealed a significant main effect of Condition (B = -0.24, SEB = 0.06, t(23.50) = -4.16, p < 0.001), as both groups showed fewer pitch interval errors in the music condition (M(SD) = 23.3(7.37)) than in the speech condition (M(SD) = 32.24(5.46)). Age was a significant predictor of the performance on pitch interval errors (B = −0.03, SEB = 0.01, t(59.10) = -2.60, p = 0.01), with older age associated with fewer interval errors. No other remaining main effects or interactions were significant. In addition, Pearson correlations confirmed the significant association between pitch interval errors and age (r(124) = −0.21, p = 0.02), but not with PPVT-R (r(124) = −0.04, p = 0.59).

Absolute duration difference

Figure 4(a) shows the distribution of the absolute duration differences for each group in both the Speech and Music conditions. The linear mixed-effects model revealed, as shown in Table 4, significant main effects of Group (B = 13.95, SEB = 4.91, t(58.81) = 2.84, p = 0.006), Condition (B = 64.65, SEB = 4.61, t(61.02) = 14.03, p < 0.001), as well as a Group * Condition interaction (B = 14.35, SEB = 4.61, t(61.02) = 3.11, p = 0.003). Post hoc analyses with Holm-Bonferroni correction for multiple comparisons suggested that both groups showed larger absolute duration differences in the music condition than in the Speech condition (Autism: t(61.1) = −12.41, p < 0.001; Non-Autism: t(60.9) = −7.54, p < 0.001, and the autism group produced larger absolute duration differences than did the non-autism group in the music condition (t(119) = −4.21, p < 0.001, Autism: M(SD) = 222.42(121.64); Non-Autism: M(SD) = 156.5(84.82)) but not in the speech condition (t(119) = 0.06, p = 0.95, Autism: M(SD) = 64.99(50.05); Non-Autism: M(SD) = 56.08(29.31)). Receptive vocabulary was a significant predictor of the performance on absolute duration matching (B = −0.8, SEB = 0.22, t(58.76) = −2.71, p = 0.009), with larger vocabulary associated with greater accuracy in absolute duration matching. The effect of Age was not significant. Again, Pearson correlations confirmed the significant association between the absolute duration differences and PPVT-R (r(124) = −0.22, p = 0.02), but not with age (r(124) = −0.02, p = 0.79).

OPEN IN VIEWER

Figure 4.

Duration-related measures for the autism and non-autism groups. (a) Absolute duration differences (in milliseconds), with black lines representing mean values. (b) Relative duration differences (in milliseconds), with black lines representing mean values. (c) Number of time errors, with error bars representing the standard deviation. Different plots are selected depending on the nature of the data type, with (a) and (b) representing continuous data, and (c) representing discrete data.

OPEN IN VIEWER

Table 4.

Coefficients for the linear mixed-effects models for duration-related measures.

Measure	Effect	Estimate	Std. Error	df	t	p
Absolute duration difference	Group	13.95	4.91	58.81	2.84	0.006**
	Condition	64.65	4.61	61.02	14.03	< 0.001***
	PPVT-R	–0.58	0.22	58.76	–2.71	0.009**
	Age	0.19	1.68	58.85	0.11	0.91
	Group × Condition	14.35	4.61	61.02	3.11	0.003**
Relative duration difference	Group	10.12	4.90	58.94	2.07	0.04*
	Condition	42.86	4.67	60.96	9.18	< 0.001***
	PPVT-R	–0.63	0.22	58.88	–2.90	0.005**
	Age	0.80	1.68	58.98	0.48	0.63
	Group × Condition	8.53	4.67	60.96	1.83	0.07
Time errors	Group	0.15	0.07	68.00	2.08	0.04*
	Condition	–0.48	0.07	69.65	–6.68	< 0.001***
	PPVT-R	–0.005	0.003	58.92	–1.67	0.10
	Age	–0.001	0.02	59.04	–0.44	0.67
	Group × Condition	0.14	0.06	61.13	2.15	0.04*

Bold values indicate statistical significance at p < 0.05. *p < 0.05, **p < 0.01, ***p < 0.001.

Relative duration difference

Figure 4(b) shows the distribution of the relative duration differences for each group in both the Speech and Music conditions. The linear mixed-effects model revealed significant main effects of Group (B = 10.12, SEB = 4.90, t(58.94) = 2.07, p = 0.04) and Condition (B = 42.86, SEB = 4.67, t(60.96) = 9.18, p < 0.001). Both groups showed larger relative duration differences in the music condition than in the speech condition, and the autism group produced larger relative duration differences than did the non-autism group not only in the music condition (Autism: M(SD) = 163.15(129.26); Non-Autism: M(SD) = 116.66(80.55)) but also in the speech condition (Autism: M(SD) = 60.64(53.99); Non-Autism: M(SD) = 48.2(30.89)). Similarly, receptive vocabulary was a significant predictor of performance on relative duration matching (B = −0.63, SEB = 0.22, t(58.88) = −2.90, p = 0.005): the larger the receptive vocabulary of the participants, the greater the accuracy in their relative duration matching. The interaction between Group and Condition, and the effect of Age were not significant (see Table 4). Pearson correlations confirmed the significant association between the relative duration differences and PPVT-R (r(124) = -0.26, p = 0.003), but not with age (r(124) = 0.007, p = 0.94).

Number of time errors

Figure 4(c) shows the distribution of the number of time errors for each group in both the Speech and Music conditions. The linear mixed-effects model revealed significant main effects of Group (B = 0.15, SEB = 0.07, t(68) = 2.08, p = 0.04) and Condition (B = -0.48, SEB = 0.07, t(69.65) = -6.68, p < 0.001). The interaction between Group and Condition was also significant (B = 0.14, SEB = 0.06, t(61.13) = 2.15, p = 0.04). Post hoc analyses with Holm-Bonferroni correction for multiple comparisons suggested that both groups showed fewer time errors in music imitation than in speech imitation (Autism: t(70.6) = 3.61, p < 0.001; Non-Autism: t(69.8) = 6.28, p < 0.001), and the autism group performed worse than the non-autism group in the music condition (t(128) = -2.98, p = 0.003, Autism: M(SD) = 33.67(16.53); Non-Autism: M(SD) = 20.7(15.31)), but not in the speech condition (t(129) = -0.12, p = 0.90, Autism: M(SD) = 46.27(11.22); Non-Autism: M(SD) = 45.03(10.63)). The effects of PPVT-R and Age did not reach significance (see Table 4).