Speech Perception and Phonological Awareness Skills in Thai-Speaking Children with Speech Sound Disorders

Abstract

Introduction

Children with speech sound disorders (SSDs) may have difficulties with speech perception and/or phonological awareness. Research on this area remains limited in children who speak tonal languages, such as Thai. This study aimed to investigate whether Thai-speaking preschool children with SSD show difficulties in speech perception and phonological awareness compared to age-matched typically developing (TD) children. The study also examined the relationships among age, speech production, vocabulary abilities, and speech perception and phonological awareness within both SSD and TD groups.

Methods

A total of 76 Thai-speaking children (SSD n = 32, TD n = 44) in kindergarten levels 2–3, aged 4;0–6;6 years, participated in this study. All children completed speech and vocabulary assessments, as well as speech perception and phonological awareness experiments. The speech perception experiment included two tasks: (1) phonemic discrimination and (2) lexical and phonetic judgement. The phonological awareness experiment included three tasks: (1) initial phoneme identification, (2) rime awareness, and (3) lexical tone awareness.

Results

Children with SSD did not perform differently from their TD peers when controlling for age and receptive vocabulary. Age predicted performance on phonemic discrimination and rime awareness in the SSD group, and on lexical and phonetic judgement and rime awareness in the TD group. Receptive vocabulary predicted only rime awareness in the SSD group. Speech production, as measured by percentage consonants correct, was not a significant predictor of performance in either group.

Conclusion

The findings provide new evidence on speech perception and phonological awareness performance in Thai-speaking children. Children with SSD did not show difficulties in speech perception or phonological awareness in this study, which may be explained by the mild-to-moderate severity of SSD in these children. This study highlights the need to develop standardised Thai articulation and phonological assessments to facilitate more precise classification of children into specific SSD subtypes in future studies.

Introduction

Children with speech sound disorders (SSDs) have difficulties producing speech sounds, resulting in reduced intelligibility [1]. They may also have problems with speech perception [2, 3] and/or phonological awareness (PA) [4, 5]. Speech perception is the process of creating a sound-based representation from acoustic inputs. These inputs can include tasks such as discrimination, identification, or recognition of spoken words [6]. PA is the ability to manipulate words into smaller units, and this skill is essential for learning to read [7, 8]. A hierarchy of levels of PA corresponds to word structure, beginning with syllable awareness, followed by onset-rime awareness and then phoneme awareness [7, 9]. Both speech perception and PA rely on phonological processing, which appears to be a problem in children with SSD [6]. Consequently, examining both speech perception and PA skills allows for a better understanding of phonological processing in children with SSD.

Speech Perception

A systematic review on speech perception skills in children with SSD [10] showed that some or all children with SSD have problems with speech perception skills, especially at preschool and early school age. This systematic review included participants speaking five languages, with English as the primary language (87.67%), followed by Portuguese, Dutch, Swedish, and Greek. For example, Edwards et al. [11] investigated speech perception skills in English-speaking preschool children applying a minimal-pair discrimination task, where the only differing phoneme was in word final position. The results showed that children with SSD obtained lower scores than their same-aged typically developing (TD) peers. Similarly, Hearnshaw et al. [2] developed a lexical and phonetic judgement based on the Canadian English version of the Speech Assessment and Interactive Learning System (SAILS) programme [12]. This task focused on the perception of the most common phonemic errors, /ʃ, s, k, ɹ/, in Australian-English-speaking children aged between 48 and 60 months. Results revealed that children with SSD showed lower performance than TD children.

Likewise, in Portuguese-speaking children aged 5;3 to 7;11, those with SSD obtained lower scores on the phonemic discrimination task than TD peers [13]. These difficulties in speech perception for children with SSD may be attributed to limited access to the perceptual representation of speech sounds [14]. Consequently, their phonological representations are not sufficient to map or facilitate the production of speech sounds [15]. As yet, the direction of the relationship between speech perception and production is unclear. Differing theoretical models predict either that speech perception impairment causes speech production difficulties [5, 16]; or that the direction of causality is that speech production impairment impacts speech perception [17, 18]. Ultimately, Hearnshaw et al. [10] reported that most research supports the theory that speech perception influences speech production. However, the level of speech perception required to assist accurate speech production requires further investigation [10]. Moreover, the task design for speech perception across studies was diverse. Hearnshaw et al. [10] also highlighted that the most common speech perception task design was lexical and/or phonetic judgement, which requires participants to judge whether a spoken word is produced correctly or incorrectly. Other frequently used tasks include identifying or discriminating minimal-pair words, in which participants are required to identify a word from a set of choices that match a spoken word or decide whether spoken words are the same or different. Although there were differences in task design, most studies found that children with SSD had difficulty with speech perception compared to TD peers [10].

Phonological Awareness

Additional difficulties that may be found in children with SSD include problems with PA and reading skills [4]. As mentioned earlier, the hierarchical levels of PA correspond to the structure of words. Children are more likely to develop rime awareness before phoneme awareness skills, and their abilities in syllable and rime awareness predict their later performance in phoneme awareness [9]. Research exploring PA skills in preschool English-speaking children with SSD used the rime matching, onset matching, and onset segmentation and matching tasks in two groups of children aged 4;0 to 4;11 (SSD n = 13, TD n = 13) [19]. The results found that children with SSD had lower scores on PA tasks compared to TD children, despite having language abilities at a normal level. Another study sought to examine the influence of PA difficulties in English-speaking children with SSD aged between 3;5 and 5;6 years [4]. This study applied varied measurements such as speech and language assessments, PA, reading, letter knowledge, and phonological representation tasks. The PA tasks consisted of elision and blending, including both receptive and expressive tasks. The results demonstrated that children with moderate SSD showed lower performance in PA tasks than TD children with the same language ability levels. Anthony et al. [4] suggested this might be due to a difficulty forming and accessing quality phonological representations.

Relationships between Age, Speech Production, Speech Perception, PA, and Vocabulary

Quality of phonological representations is likely to also effect the ability to learn words. Researchers are therefore also interested in the relationships between age, speech production, speech perception, PA, and vocabulary [5, 11, 20]. Edwards et al. [11] found that size of receptive vocabulary, but surprisingly not age was a significant predictor in this study of speech perception skills. Moreover, Hearnshaw et al. [21] found that speech production and both expressive and receptive vocabulary were significant predictors of speech perception skills in 61 Australian English-speaking children aged 48–69 months with and without SSD. Hearnshaw et al. [21] suggested that enhancing vocabulary skills may improve the quality of phonological representation, as children who scored highly on vocabulary were more likely to succeed on the speech perception task. Likewise, Rvachew and Grawburg [5] examined the relationship between PA skills, speech perception, and receptive vocabulary in preschoolers with SSD. They found that speech perception tended to directly influence PA and that receptive vocabulary mediated the effect of speech perception on PA. For older children aged 7;0 to 17;4, Benway et al. [22] found that speech perception and vocabulary directly influenced PA. In contrast to preschoolers, vocabulary did not mediate the relationship between speech perception and PA in school-aged children [22]. Preston and Edwards [20] studied predictors of PA in 43 preschoolers with SSD, and found that age and receptive vocabulary impacted 33% of the variance in PA, and atypical speech errors predicted 6% of the variance in PA. Preston and Edwards [20] suggested that phonological representation in children with SSD could be impaired. Phonological representation therefore appears to develop alongside vocabulary as children grow [20, 23].

Speech Perception and PA in Thai-Speaking Children

The above studies are mostly in English or in other non-tonal languages. There has been limited research on speech perception and PA in children with SSD who speak tonal languages, such as Thai. Tonal languages differ from non-tonal languages because contrastive pitch occurs across all levels of the phonological hierarchy. In contrast, contrastive pitch cannot be found at the segment level in non-tonal languages. Non-tonal language speakers tend to perceive lexical tones as paralinguistic information, such as stress or prosody contrast [24]. Therefore, studies in the Thai population provide data on speech perception and PA of word- and segment-level pitch contrast (lexical tone), which would be missing in a study of non-tonal languages.

Recently, a scoping review on PA skills in Thai-speaking children found that no studies focused on children with SSD [25]. Of the 13 studies included in the review, most focused on PA skills in TD children or children at risk of dyslexia or learning disabilities. Of the 13 studies, the most common PA task used was initial phoneme identification. Given that Thai is a tonal language, a lexical tone awareness task was also included in a small number of studies [26, 27].

In terms of speech perception in Thai, previous studies employed auditory discrimination (using minimal pairs) to examine speech perception skills in children with SSD aged between 5;0–7;11 years [28] and 3;0–6;11 years [29]. Jeeraumporn [28] included minimal pairs of both real words and non-words that differed in consonants, vowels, or tones. Children were asked to judge whether the words were the same or different. In contrast, Arayathaweekun [29] included only minimal pairs of real words that differed in consonants or vowels. Children were asked to identify the picture of the word spoken by the tester. The results from both of these studies indicated that TD children performed better than children with SSD in the auditory discrimination tasks. Moreover, Jeeraumporn [28] also reported that auditory discrimination skills in children with SSD were related to their speech production, although this relationship was not found by Arayathaweekun [29].

Task Rationale for the Current Study

The purpose of speech perception tasks is to assess children’s perceptual (acoustic-phonetic) representations and determine how accurately they function. These tasks require children to access perceptual representations and match them with spoken words [6, 30]. Locke [30] suggested that tasks requiring children to say whether two syllables or words are the same or different cannot determine whether children actually access their perceptual representations. Moreover, the concepts of “same” and “different” may be too complex for children to understand [6]. On the other hand, tasks that require children to make perceptual decisions, such as pointing to the word they hear from a set of minimal pairs or judging whether a spoken word is correctly produced, are more likely to require children to compare spoken stimuli with their perceptual representations [30]. Therefore, the current study included a phonemic discrimination task, in which children were presented with a two-picture forced choice of minimal pairs and asked to select the picture corresponding to the spoken stimulus. The other speech perception task was a lexical and phonetic judgement task, in which children judged whether the spoken stimulus was correctly or incorrectly produced. This task was expected to require more from children’s perceptual representations than the phonemic discrimination task, as it expected them to access lexical knowledge and evaluate the accuracy of phonetic detail. These two tasks have been used in studies of children aged 36–83 months [2, 29], so we considered them appropriate for the age group in the current study.

PA tasks can vary in difficulty, so task selection should consider both children’s age and their level of PA skills [6]. Assessing PA skills in preschool children may be challenging; therefore, tasks should focus on larger phonological units, e.g., syllables or onset-rime units [31]. A scoping review of PA in Thai-speaking children [25] reported that commonly used tasks for preschoolers included initial phoneme identification and lexical tone awareness. However, rime awareness tasks were not included in any studies with this age group. Therefore, the current study included three PA tasks, initial phoneme identification, rime awareness, and lexical tone awareness, to examine children’s PA abilities at both the onset-rime level and the level of lexical tones. To facilitate a better understanding of these tasks in the Thai version, the following section provides information on Thai segments and phonotactic constraints.

Segments and Phonotactic Constraints of Thai Phonology

Thai has twenty-one initial consonant phonemes and nine final-consonant phonemes [32]. It has eighteen monophthongs (nine short vowels and nine long vowels), and three diphthongs. Thai’s five tones are mid, low, rising, high, and falling [32]. Thai words are mostly monosyllabic, in which the syllable structure includes onset, nucleus, coda (optional), and tone [33]. Consonant clusters appear only in the prevocalic position; therefore, the coda position can only have one phoneme [33]. The Thai writing system is an abugida, where consonant letters represent syllables with inherent vowels, /o/ in closed syllables and /a/ in open syllables, while diacritics are used to indicate tones and other phonetic features [34]. Thai has 44 consonants and 32 vowel letters, as well as four tone diacritics [35].

In sum, previous literature suggests that children with SSD are more likely to have difficulties with speech perception and PA. Previous studies have also examined the relationships among age, speech production, vocabulary skills, and both speech perception and PA, with varying results reported depending on the characteristics of the participants or the task design.

Specifically, in the Thai context, studies on this topic remain limited. Although studies on speech perception in Thai-speaking children with SSD [28, 29] examined group differences, the relationships among factors, such as speech perception and vocabulary skills, still require further investigation. Moreover, no studies investigated PA skills in Thai-speaking children with SSD, as most previous research on PA in Thai focused on children at risk of dyslexia or learning disabilities [25]. Therefore, in the present study, we sought to contribute to theoretical understanding by providing preliminary evidence on whether Thai-speaking children with SSD have difficulties with speech perception and PA. This extends existing knowledge by examining children’s abilities across multiple levels of speech perception and PA tasks. In terms of clinical practice, we aimed to highlight the importance of more comprehensive assessment and intervention for Thai-speaking children with SSD. Clinicians may need to consider not only speech production but also speech perception, PA, and vocabulary skills in clinical assessment and intervention. We therefore examined the relationships among age, speech production, vocabulary, speech perception, and PA. Age, speech production, and vocabulary were selected as they have been identified as potentially important predictors in previous studies [11, 20, 22]. Sex was not included as a predictor in the current study, as it was not included in the analytical models of previous studies on similar topics [20–22]. Nevertheless, zero-order relationship between sex and both speech perception and PA scores was examined. The research questions are as follows:

• 1.Do Thai-speaking children with SSD have lower scores than TD children in speech perception and PA tasks?
• 2.Are age, speech production, and vocabulary abilities related to speech perception and PA within each group?

Methods

Participants

Seventy-six children aged 48–78 months participated in this study, comprising 32 children with SSD (M = 59.84, SD = 6.63) and 44 TD children (M = 62.48, SD = 7.13). The mean ages between the two groups were not significantly different (t(74) = −1.64, p = 0.11). The children were in kindergarten levels 2 to 3. The SSD group comprised 15 girls and 17 boys, while the TD group comprised 25 girls and 19 boys. All 44 TD children were recruited from two kindergartens and a primary school with kindergarten in Bangkok and its surrounding provinces. Seven children with SSD were recruited from two hospitals: Ramathibodi Hospital, Bangkok, and Ramathibodi Chakri Naruebodindra Hospital, Samut Prakan. The remaining 25 children with SSD were recruited from the schools mentioned above, as they were found to have SSD during the speech assessment prior to the main experiment (see below).

All participants were monolingual Thai-speaking children who speak the Central Thai dialect and had normal hearing. In the SSD group, the inclusion criteria included children who were diagnosed with SSD by speech and language therapists (SLTs) (including the first author), had receptive and expressive language development within the normal to mild delay range, and had no reported intellectual disability. Children who were diagnosed with motor or neurological disorders, cleft palate with or without cleft lip or other craniofacial differences, or known genetic syndromes affecting speech, language, or cognition were excluded. The diagnostic procedure used by the first author followed the same protocol as that employed by the SLTs at the hospital. We used the Thai Articulation Test, adapted from the articulation test by Boonyathitisuk [36], and compared the results with the Thai speech acquisition tables of Mukngoen [37] and Boonyathitisuk [36]. Determining SSD involved comparing speech errors against typical children’s sound acquisition based on Mukngoen [37] and Boonyathitisuk [36]. Children were defined as having SSD if they produced errors on sounds typically acquired by at least 75% of children at that age. Although no formal distinction between delay and disorder was made, error patterns were considered in relation to whether they aligned with typical developmental substitution/simplification described in Mukngoen [37] and Boonyathitisuk [36], or were atypical and/or not consistent with developmental patterns reported in Mukngoen [37] and Boonyathitisuk [36]. The first author is a practising SLT working in the same service as the SLTs who assisted with study recruitment; thus, there was consistency with routine clinical practice. The inclusion criteria in the TD group included children who had normal receptive and expressive language development, no reported intellectual disability, and no history of SSDs or specific learning difficulties, such as dyslexia.

Speech and Language Assessments

All participants underwent a battery of speech and language assessments to confirm eligibility. There are currently no standardised articulation or phonology assessments in Thai. The children’s speech productions were assessed using the Thai Articulation Test, adapted from the articulation test by Boonyathitisuk [36], a picture-naming task. The test consists of 96 pictures: 63 designed to elicit all 21 initial consonants, 22 designed to elicit eight final consonants, and 11 designed to elicit 11 consonant clusters. All five tones were represented across the 96 pictures. TD children in this study were expected to produce all tones, final consonants, and most initial consonants correctly, except for four initial sounds (/f/, /tɕʰ/, /s/, /r/) and consonant clusters, as these are still developing between ages 4;0 and 7;0 [36, 37]. Children were assigned to the SSD group if they presented with errors on sounds that are expected to be correct at their individual age, e.g., final-consonant errors [36, 37]. Following the speech assessment, each participant was evaluated for their receptive and expressive vocabulary skills. There are no standardised comprehensive language assessments available in Thai; hence, a test focusing on vocabulary was employed: the Thai Semantic Development Test [38]. This test is a standardised test that comprises three subtests: the receptive vocabulary subtest (RV) is the picture vocabulary, while the two expressive vocabulary subtests (expressive vocabulary subtest 1 [EV1] and expressive vocabulary subtest 2 [EV2]) are the relational vocabulary and oral vocabulary, respectively. The picture vocabulary subtest measures the understanding of word meaning, while the relational vocabulary subtest measures the expression of the relationship between two words. The oral vocabulary subtest measures the ability to explain familiar words. In addition to the battery of speech and language assessments, the Intelligibility in Context Scale (ICS) [39] was distributed to parents or carers to assess their child’s speech intelligibility over the preceding month.

Speech Perception and PA Experiments

The speech perception experiment included two tasks: (1) phonemic discrimination (PD) and (2) lexical and phonetic judgement tasks (LPJ). The PA experiment consisted of three tasks: (1) initial phoneme identification (IPI), (2) rime awareness (RA), and (3) lexical tone awareness (LTA) tasks. All tasks were created using the Gorilla Experiment Builder (www.gorilla.sc) [40], a cloud-based platform. Picture stimuli were obtained from Canva, Pixabay, and Freepik. Speech stimuli were audio-recorded from a 30-year-old Thai-speaking female who speaks the central Thai dialect. The details of each task are described below.

Speech Perception Experiment

Phonemic Discrimination Task (15 Items)

This task was a two-picture forced choice: the target and a distractor. Each test item was a minimal pair of monosyllabic words with one feature difference on either the initial or final phoneme. Some minimal pairs in this task were taken from previous studies on auditory discrimination abilities in Thai-speaking children [29, 41]. Additional minimal pairs were added by the author. In total, this task included 15 minimal pairs (30 words), such as contrasts between /s/ – /t/ and /ŋ/ – /n/.

For example, the target stimulus was /mǔː/ pig presented aloud, and the two-picture forced choice was /mǔː/ pig and /nǔː/ rat (see Fig. 1a). The contrast between the target and distractor was therefore in the place of articulation (bilabial versus alveolar). The child had to choose the picture of a pig for the correct response.

Fig. 1.

a–e Example of experimental screens across five tasks. PD, phonemic discrimination; LPJ, lexical phonetic judgement; IPI, initial phoneme identification; RA, rime awareness; LTA, lexical tone awareness.

Lexical and Phonetic Judgement Task (15 Items)

This task was adapted from the SAILS programme designed for Canadian English-speaking children [12]. Each child listened to a stimulus and judged whether it was a correct or incorrect production of the picture stimulus shown. If the speech stimulus was a correct production, the child had to select a picture of a tick. In contrast, if it was an incorrect production, the child had to select a picture of a cross. The incorrect productions were designed based on phonological processes that might be found in TD Thai-speaking children, such as stopping, deaspiration, gliding, fronting, backing, final-consonant deletion, and devoicing [42]. In total, this task included 15 words: 7 correctly produced and 8 incorrectly produced.

For example, the picture of the fork was displayed, and each child listened to the speech stimulus /s$\widehat{\mathrm{ɔ}}$ːm/ fork (correct production). The child had to point at a picture of the tick for the correct response. However, if the stimulus was /t$\widehat{\mathrm{ɔ}}$ːm/ (incorrect production, where the phonological process was stopping), the child had to point at the picture of the cross (see Fig. 1b).

PA Experiment

Initial Phoneme Identification Task (10 Items)

This task was adapted from the PA task in the Rama Pre-Read software programme designed for Thai-speaking children [43]. Each item consisted of four pictures: stimulus, target, and two distractors. The child was asked to listen to a speech stimulus and identify which picture had a word containing the same initial phoneme as the stimulus. Overall, the task included 40 words: 10 stimulus words, 10 target words, and 20 distractors. The stimuli comprised the following initial phonemes: /n/, /t/, /m/, /kʰ/, /s/, /b/, /tɕʰ/, /h/, /pʰ/, and /f/.

For example, a stimulus for IPI was /kʰàj/ egg. A child was shown a stimulus picture (egg) and three picture choices: /kʰ$\check{\mathrm{a}}$ː/ leg, /tɕʰ$\widehat{\mathrm{ɯ}}$ak/ rope, and /n$\overline{\mathrm{o}}$m/ milk. Then, the child listened to the speech stimulus and was asked to identify which choice had the same initial phoneme as /kʰ$\stackrel{`}{\mathrm{a}}$j/ egg (see Fig. 1c). The correct response was /kʰ$\check{\mathrm{a}}$ː/ leg in this case.

Rime Awareness Task (10 Items)

Each child was shown four pictures for each item: stimulus, target, and two distractors. Then, the child listened to a speech stimulus and was asked to select a picture that contained the word with the same rime as the speech stimulus. Overall, the task included 40 words: 10 stimulus words, 10 target words, and 20 distractors, examples included /l$\check{\mathrm{a}}$ŋ/ – /tʰ$\check{\mathrm{a}}$ŋ/ and /p$\overline{\mathrm{u}}$ː/ – /ŋ$\overline{\mathrm{u}}$ː/.

For example, a child was shown a picture of /m$\check{\mathrm{i}}$ː/ bear as a stimulus, and there were three picture choices: /w$\check{\mathrm{i}}$ː/ comb, /t$\mathrm{ó}$ʔ/ table, and /k$\widehat{\mathrm{\varepsilon }}$ːw/ glass. Each child listened to the speech stimulus and then selected a picture with the same rime as /m$\check{\mathrm{i}}$ː/ bear (see Fig. 1d). The correct answer for this case was /w$\check{\mathrm{i}}$ː/ comb.

Lexical Tone Awareness Task (15 Items)

The word list for this task was adapted from the study on the perception of Thai lexical tone in Thai-speaking children [26]. Each item was a two-picture forced choice of a minimal pair with a tone difference. In total, this task included 15 minimal pairs (30 words), such as contrasts between low and mid tones and between rising and falling tones.

For example, there were two pictures: /s$\check{\mathrm{ɯ}}$a/ tiger and /s$\widehat{\mathrm{ɯ}}$a/shirt. The speech stimulus was /s$\check{\mathrm{ɯ}}$a/ tiger, and the child listened to the stimulus and chose the picture of the tiger for the correct response (see Fig. 1e).

Data Collection Procedure

Data collection was completed in one session lasting 60–70 min with breaks as required. Firstly, each child was assessed for their speech production using the Thai Articulation Test adapted from Boonyathitisuk’s Articulation Test [28]. Secondly, the child was assessed on their vocabulary skills using the Thai Semantic Development Test [38]. The last part consisted of the experimental tasks: the two speech perception tasks and three PA tasks described above. The tasks were ordered from earliest acquired to later acquired, i.e., potentially increasing in difficulty. The child was given a short break between tasks to reduce fatigue, maintain attention, and mitigate potential order effects. The sequence of the tasks was (1) PD, (2) LTA, (3) LPJ, (4) RA, and (5) IPI. Each child was asked to name all the pictures before starting the experiment to ensure they recognised the pictures and vocabulary. There were three practice trials in each task, in which the researcher supported the child via demonstrating and modelling if need be. Each child listened to the speech stimuli through external speakers. Picture stimuli for each task were shown on a laptop screen via the experiment platform Gorilla. The child responded to speech stimuli by pointing to a picture on the laptop screen, which the researcher then selected. Reaction time was not collected in this study. The repetition of items was allowed only once per item. One point was given for each correct response. All children were presented with all items, although some of them struggled with several items.

Reliability and Validity

Inter-rater reliability of transcription from the Thai Articulation Test, adapted from the articulation test by Boonyathitisuk [36], was assessed using point-to-point agreement on a random selection of 20% of the data (16 children). The first author and another Thai SLT independently transcribed 202 points for consonants, 111 points for vowels, and 111 points for lexical tones in each participant. Inter-rater reliability agreement was 95.79% for consonants, 99.32% for vowels, and 99.38% for lexical tones. While specific psychometric data for validation of this adapted articulation test are not available, this test was developed following principles of articulation assessment used by Boonyathitisuk [36], ensuring that the words are designed to elicit the entire Thai consonant inventory (in front, central, and back vowel contexts for each consonant), the full Thai lexical tone inventory, and every consonant cluster.

The current study was a pilot study to explore the feasibility and usability of the speech perception and PA tasks. Task content was designed based on theoretical principles and reviewed by the second and fourth authors to support content validity. Intra-rater reliability for correct and incorrect responses across five tasks, assessed by the first author, was 100%. All responses from the speech perception and PA tasks were recorded and exported from the Gorilla experiment platform.

Data Analysis

Statistical analyses were performed using R version 4.5.0 [44]. Group comparisons of performance across five tasks were performed using five separate generalised linear mixed models (GLMMs). The dependent variables were each of the task scores. The effect of Group was included as a fixed effect to examine its influence on the dependent variables. Given the repeated observations nested within each child and varying levels of item difficulty, the effects of Child and Stimulus were included as random effects. The binomial family was selected for the GLMMs because the response was binary (1 = correct, 0 = incorrect). To consider covariates in the models, the variance inflation factor (VIF) and correlation analyses were performed on age, speech production scores, and vocabulary scores. These analyses were used to check potential multicollinearity among covariates. Moreover, separate GLMMs were conducted within each group to examine how age, speech production, and vocabulary skills relate to performance in speech perception and PA tasks. Point-biserial correlation was used to examine the zero-order relationship between sex and both speech perception and PA scores.

In addition to statistical analyses, speech error patterns were examined to characterise the speech profiles of children with SSD. Error patterns were analysed into phonological patterns (e.g., velar fronting, gliding, cluster reduction, stopping) or phonetic patterns (e.g., articulatory distortion of fricative /s/, lateralisation). Error patterns occurring three or more times in each child were counted. In the current study, we did not classify children with SSD into subtypes, as no standardised Thai articulation and phonology assessments are currently available. There is no universal classification system for SSDs [45]. In Thailand, SSDs are not classified into subtypes; instead, all cases are diagnosed as specific speech articulation disorder (F80.0) according to the International Classification of Diseases and Related Health Problems (ICD-10) [46].

Results

The descriptive statistics for age, percentage consonants correct (PCC), and vocabulary scores from the Thai Semantic Development Test [38] are shown in Table 1. PCC was calculated from speech productions in the Thai Articulation Test, using the total accurate consonants divided by the total number of consonants, multiplied by 100 [47]. A Mann-Whitney U test showed a significant difference in PCC between SSD and TD groups (U = 0, p < 0.001). The median (IQR) PCC was 76.73 (71.29–85.89) in the SSD group and 93.07 (93.07–93.19) in the TD group. The results of PCC in the SSD group (M = 76.72, SD = 10.01) indicated that the severity of SSDs falls within the mild-to-moderate range based on the PCC severity scale [47]. In contrast, the mean PCC for the TD group was 93.01 (SD = 1.03), which confirms that TD children had more accurate speech production than those with SSD. This is in line with the hypothesis that children with SSD produce more speech errors than TD children.

Table 1.

Summary of participant profiles

Participant profile	Group
	SSD mean (SD)	TD mean (SD)
Age, months	59.84 (6.63)	62.48 (7.13)
Thai articulation test (PCC)	76.72 (10.01)	93.01 (1.03)
Thai semantic development test
RV subtest (z-score)	0.02 (0.73)	0.29 (0.53)
EV1 (z-score)	0.38 (0.96)	0.74 (0.64)
EV2 (z-score)	0.05 (1.14)	0.68 (0.64)

The mean of each characteristic is shown with the standard deviation in brackets.

SSD, speech sound disorder; TD, typically developing; PCC, percentage consonants correct; RV, receptive vocabulary; EV, expressive vocabulary.

Regarding vocabulary skills, the cut-off for delayed vocabulary development was defined as z-scores below −1.25 SD [48]. The results showed that no participants had RV delay, with mean z-scores of 0.02 in the SSD group and 0.29 in the TD group. However, in the SSD group, one participant had a z-score of −1.27 on the EV1 subtest, and six participants had z-scores ranging from −1.43 to −1.27 on the EV2 subtest, suggesting borderline EV delay.

The results of ICS [39] in the TD group (see Table 2) indicated that parents rated other acquaintances¹ as understanding their children’s speech the best (M = 4.59, SD = 0.54). On average, parents rated themselves and their child’s friends as equally understanding of their children’s speech (M = 4.50, SD = 0.51). Strangers understood their children’s speech the least (M = 4.25, SD = 0.58). On the other hand, parents in the SSD group (see Table 3) rated themselves as understanding their children’s speech the best (M = 4.22, SD = 0.66), followed by immediate family members (M = 4.06, SD = 0.76) and acquaintances (M = 4.00, SD = 0.62). Their children’s friends and teachers showed equal understanding of their children’s speech (M = 3.78, SD = 0.61), while strangers were the group with the lowest rating (M = 3.44, SD = 0.62).

Table 2.

ICS scores in the TD group

​	Questions	Mean	SD	Range	Always	Usually	Sometime	Rarely	Never
1	Do you understand your child?	4.50	0.51	4–5	50.00	50.00	0	0	0
2	Do immediate members of your family understand your child?	4.50	0.55	3–5	52.27	45.45	2.27	0	0
3	Do extended members of your family understand your child?	4.27	0.82	1–5	43.18	45.45	9.09	0	2.27
4	Do your child’s friends understand your child?	4.50	0.51	4–5	50.00	50.00	0	0	0
5	Do other acquaintances understand your child?	4.59	0.54	3–5	61.36	36.36	2.27	0	0
6	Do your child’s teachers understand your child?	4.41	0.62	3–5	47.73	45.45	6.82	0	0
7	Do strangers understand your child?	4.25	0.58	3–5	31.82	61.36	6.82	0	0
​	Average total score (maximum = 35)	31.02	3.27	25–35	​	​	​	​	​
​	Average total score (maximum = 5)	4.43	0.47	​	​	​	​	​	​

Table 3.

ICS scores in the SSD group

​	Questions	Mean	SD	Range	Always	Usually	Sometime	Rarely	Never
1	Do you understand your child?	4.22	0.66	3–5	34.38	53.13	12.50	0	0
2	Do immediate members of your family understand your child?	4.06	0.76	3–5	31.25	43.75	25.00	0	0
3	Do extended members of your family understand your child?	3.63	0.66	2–5	6.25	53.13	37.50	3.13	0
4	Do your child’s friends understand your child?	3.78	0.61	3–5	9.38	59.38	31.25	0	0
5	Do other acquaintances understand your child?	4.00	0.62	3–5	18.75	62.50	18.75	0	0
6	Do your child’s teachers understand your child?	3.78	0.61	3–5	9.38	59.38	31.25	0	0
7	Do strangers understand your child?	3.44	0.62	2–5	3.13	40.63	53.13	3.13	0
​	Average total score (maximum = 35)	26.91	3.70	20–35	​	​	​	​	​
​	Average total score (maximum = 5)	3.84	0.53	​	​	​	​	​	​

Research Question 1: Group Comparison of Performance in Speech Perception and PA

The percentage scores from the five tasks across the SSD and TD groups are illustrated in Figure 2. The total scores for the two speech perception tasks, PD and LPJ, were 15 points. For the three PA tasks, RA and IPI each had a total score of 10 points, whereas LTA had a total score of 15 points.

Fig. 2.

Box plots of percentage scores for the SSD and TD groups across five tasks. SSD, speech sound disorder; TD, typically developing; IPI, initial phoneme identification; LPJ, lexical phonetic judgement; LTA, lexical tone awareness; PD, phonemic discrimination; RA, rime awareness.

Five GLMMs (one for each task) were performed to examine group differences between the SSD and TD groups. Age, PCC, and vocabulary scores (z-scores) were considered as covariates in five models, as we sought to identify relationships between these predictors and speech perception and PA. The VIF was calculated to check multicollinearity, and the results showed that the values of VIF among predictors were lower than four, indicating a low likelihood of multicollinearity (VIF of age = 1.25, VIF of PCC = 1.35, VIF of RV = 1.47, VIF of EV1 = 1.70, and VIF of EV2 = 2.23). However, VIF does not directly measure the relationship among predictors, but rather assesses potential multicollinearity. Therefore, correlation analysis was also conducted to examine the relationships between predictors. The results (see Table 4) showed that EV1 and EV2 were very strongly correlated (r = 0.63), and RV was moderately correlated with both EV1 (r = 0.42) and EV2 (r = 0.55). These findings suggest an overlap between EL1 and EV2, as well as overlaps between RV and both EV1 and EV2. As a result, the model included only RV as the measure of vocabulary skills to avoid redundancy. Moreover, RV was prioritised over EV, as most previous studies [5, 11, 20] suggest that RV is a predictor for speech perception and PA skills in children with SSD. The zero-order relationship between sex and task scores was examined using point-biserial correlations. The results showed that sex was not significantly associated with any of the task scores: PD, r_pb = 0.12, p = 0.30; LPJ, r_pb = −0.01, p = 0.91; RA, r_pb = −0.13, p = 0.25; IPI, r_pb = 0.04, p = 0.73; LTA, r_pb = −0.02, p = 0.86. This suggested that sex had no significant zero-order relationship with the task scores and therefore was not included as a covariate in the model.

Table 4.

Correlation among predictors

​	Predictors	1	2	3	4	5
1	Age	–	​	​	​	​
2	PCC	0.32	–	​	​	​
3	RV (z-score)	0.01	0.14	–	​	​
4	EV1 (z-score)	−0.10	0.26	0.42	–	​
5	EV2 (z-score)	−0.16	0.32	0.55	0.63	–

PCC, percentage consonants correct; RV, receptive vocabulary; EV, expressive vocabulary.

Age, PCC, and RV were initially included as covariates in all models. However, across all five models, PCC showed a strong correlation with Group in the fixed effects, with correlation values ranging from 0.73 to 0.78. Therefore, PCC was excluded from the five GLMMs used to test group differences. This was done because PCC was one of the measures used to identify children with SSD. Children with SSD showed lower PCC, so the Group effect of SSD may partly reflect differences in PCC. Each of these models included Score as the dependent variable, Group as the fixed effect, Age and RV (z-score) as covariates, and Child and Stimulus as random effects. The TD group was set as the baseline group in the models. The results of the fixed and random effects across the five GLMMs can be found in online supplementary Table A (for all online suppl. material, see https://doi.org/10.1159/000552670).

The results of five GLMMs indicated that there were no significant differences between groups across five tasks when controlling for age and RV: PD (β = −0.14, OR = 0.87, SE = 0.22, p = 0.52), LPJ (β = −0.43, OR = 0.65, SE = 0.29, p = 0.14), RA (β = 0.12, OR = 1.13, SE = 0.23, p = 0.60), IPI (β = 0.01, OR = 1.01, SE = 0.19, p = 0.94), LTA (β = −0.07, OR = 0.94, SE = 0.26, p = 0.80) (see online suppl. Table A). When focusing on the unconditional GLMM for the LPJ task (which did not include Age or RV), this was the only model in which results showed that children with SSD had significantly lower correct scores than TD children (β = −0.74, OR = 0.47, SE = 0.30, p = 0.01). However, after adding Age and RV as covariates (conditional model), the GLMM for the LPJ task no longer showed a significant group effect, as reported earlier. The unconditional model shows the raw group difference, while the conditional model accounts for potential confounding factors such as Age and RV.

Research Question 2: How Do Age, Speech Production, and Vocabulary Skills Relate to Speech Perception and PA in Each Group

Ten GLMMs (one per task per group) were fitted to examine the effects of age, PCC, and vocabulary on speech perception and PA within each group. Each model included Score as the dependent variable; Age, PCC, and RV as fixed effects; and Child and Stimulus as random effects. Predicted probabilities of correct responses were calculated from each model to visualise the data. See online supplementary Tables B and C for the GLMM results in the SSD and TD groups.

Phonemic Discrimination

In the SSD group, age was the only significant predictor with a positive effect on correct responses (β = 0.05, OR = 1.05, SE = 0.03, p = 0.04), while PCC (p = 0.90) and RV (p = 0.27) were not significant predictors. The predicted probability of correct responses across age in the SSD group was visualised in Figure 3a. In the TD group, all predictors were non-significant.

Fig. 3.

a–e Effect of age on the probability of correct responses across tasks. PD, phonemic discrimination; LPJ, lexical phonetic judgement; IPI, initial phoneme identification; RA, rime awareness; LTA, lexical tone awareness; SSD, speech sound disorder; TD, typically developing.

Lexical Phonetic Judgement

None of the predictors were significant in the SSD group. In the TD group, age was the only significant predictor with a positive effect on correct responses (β = 0.07, OR = 1.07, SE = 0.03, p = 0.03), while PCC (p = 0.86) and RV (p = 0.14) were not significant predictors. The predicted probability of correct responses across age in the TD group was visualised in Figure 3b.

Initial Phoneme Identification

None of the predictors were significant in the SSD group. The initial IPI model in the TD group produced a boundary (singular) fit warning, indicating that the random-effects structure required adjustment. The random intercept for Child was therefore removed because its variance was close to zero (<0.001). After removing this random effect, the revised model still showed convergence warnings. These warnings were likely due to the limited variability of scores in the TD group (see Fig. 3c).

Rime Awareness

In the SSD group, age and RV were significant predictors with a positive effect on correct responses (age: β = 0.06, OR = 1.06, SE = 0.03, p = 0.03; RV: β = 0.51, OR = 1.66, SE = 0.23, p = 0.03), while PCC was not a significant predictor (p = 0.43). In the TD group, age was the only significant predictor with a positive effect on correct responses (β = 0.07, OR = 1.07, SE = 0.03, p < 0.01), while PCC (p = 0.53) and RV (p = 0.51) were not significant predictors. The predicted probability of correct responses across age in SSD and TD groups was visualised in Figure 3d. Predicted probability across RV in the SSD group was visualised in Figure 4.

Fig. 4.

Effect of RV on the probability of correct responses in the RA task (SSD group).

Lexical Tone Awareness

None of the predictors were significant in the SSD group. The LTA model in the TD group showed convergence warnings, possibly due to a ceiling effect (see Fig. 3e).

In summary, the findings showed that the performance of children with SSD did not differ significantly from that of TD children across all five tasks when controlling for age and RV. Age was a significant predictor of performance on LPJ and RA tasks in the TD group and on PD and RA tasks in the SSD group. RV skill was a significant predictor only on the RA task in the SSD group. PCC was not a significant predictor on any tasks across both groups.

Phonological and Phonetic Patterns

Cluster reduction was the most frequent phonological process observed in all children in the SSD group. The next most common phonological process was velar fronting, observed in 18 of 32 participants (56.25%). Some other common phonological processes found were backing (7/32, 21.88%), devoicing (7/32, 21.88%), glottal replacement (6/32, 18.75%), stopping of affricates (5/32, 15.63%), and gliding (5/32, 15.63%). The remaining phonological processes were observed in fewer participants: medial consonant deletion (4/32, 12.5%), final-consonant deletion (3/32, 9.38%), affrication (3/32, 9.38%), deaffrication (3/32, 9.38%), stopping of fricatives (2/32, 6.25%), assimilation (2/32, 6.25%), deaspiration (2/32, 6.25%), and stopping of liquids (2/32, 6.25%). In addition to the phonological processes identified, distortions were observed in four participants for the fricative /s/ (12.5%), such as lateralisation and dentalisation, and in three participants for the affricate /tɕʰ/ (9.38%), such as fricativisation. Moreover, a substitution of /r/ with [l] was found in 25 participants (78.13%). It should be noted that some phonological processes, e.g., backing and glottal replacement, are considered atypical in English-speaking children [1]. However, no formal studies on Thai phonological processes were conducted [42]. Our analysis therefore followed Lorwatanapongsa and Maroonroge’s [42] generalised overview of phonological processes that might be observed in TD Thai-speaking children from Mukngoen [37] and Boonyathitisuk [36], including cluster reduction, velar fronting, backing, glottal replacement, stopping, labialisation, deaspiration, and stridency deletion.

Discussion

This study sought to determine whether Thai-speaking preschool children with SSD have difficulties with speech perception and PA compared to age-matched TD children. Contrary to our hypotheses, and in conflict with previous research on children with SSD speaking other languages, we found that children with SSD did not perform differently from TD children on speech perception and PA tasks when controlling for age and receptive vocabulary. The details of their performance are discussed in the sections below.

Children’s Speech and Language Profiles

The mean PCC in the SSD group indicated that children with SSD in the current study had mild-to-moderate severity SSD, as indicated using the PCC severity scale [47]. Children with SSD in the present study performed variably on speech perception and PA tasks, as most children had a combination of delayed and disordered patterns [49]. Dodd [50] suggested that different error types in SSD are associated with distinct underlying deficits. Chien et al. [51] found that children with SSD who produced atypical speech errors were more likely to have speech perception problems. Moreover, Preston and Edwards [20] suggested that PA difficulties in children with SSD were related to atypical speech errors. However, to our knowledge, there are no published studies on atypical errors among Thai-speaking children with SSD [42] so we do not yet know what these errors were. We generalised from what sound errors did not appear as developmental patterns in Mukngoen [37] and Boonyathitisuk [36]. No children showed a significant vocabulary delay in this study. Metsala [23] suggested that phonological representation would develop more accurately with vocabulary growth.

Speech Perception

To test phonemic discrimination, we used a two-picture forced choice minimal-pair task similar to the Thai auditory discrimination task in Arayathaweekun’s [29] study. Arayathaweekun’s [29] study found that TD children performed better than children with SSD (aged 3;0–6;11 years) in the auditory discrimination task, whereas we found no difference in group performance. In contrast, a study by Locke and Kutz [52] revealed that English-speaking kindergarten children who substituted [w] for /r/ performed as well as their TD peers in the speech perception task where children were asked to identify a word given by a tester from a set of six pictures (ring, wing, king, shack, sack, and tack). Locke and Kutz [52] suggested that recognition difficulties were not found in this task as compared to tasks that require children to identify whether two spoken syllables are linguistically the same or different. However, unlike the children in Locke and Kutz’s [52] study, children with SSD in our study did not exhibit only a gliding pattern. Our findings showed no difference between groups, which may be due to the mild-to-moderate severity of SSD and the heterogeneity of articulatory and phonological errors.

The lexical and phonetic judgement task was the most common task used in previous speech perception studies in SSD [10]. A meta-analysis of eight studies using the lexical and phonetic judgement task revealed that TD children obtained higher scores than children with SSD [10]. However, a recent study by Hearnshaw et al. [21] using the lexical and phonetic judgement task found no significant difference between the SSD and TD groups. They suggested that the reasons may include the following: children with SSD in their study had mild or moderate severity of SSD; many children with SSD exhibited advanced vocabulary skills; and children with SSD presented with both articulation and phonological errors. In our study, children with SSD performed significantly lower than TD children in the lexical and phonetic judgement task when age and receptive vocabulary were not controlled. However, no significant group difference was observed when age and receptive vocabulary were controlled for, nor when only age was controlled. Therefore, this could indicate that the group difference was explained by age and receptive vocabulary rather than the effect of SSD per se. However, other research suggests that receptive vocabulary may be vulnerable in children with SSD; hence, lower receptive vocabulary may reflect an effect of SSD [20, 22]. The present study showed outcomes that differed from some previous studies, which may be attributed to differences in methodology. Hearnshaw et al. [10] highlighted that the disparate methodologies used in speech perception studies of children with SSD could be the reason for the diverse findings, for example, variations in task types (discrimination or lexical and phonetic judgement tasks), stimulus types (tester voice, children’s own voice, or synthesised speech), and SSD subgroups (articulation or phonological impairment).

Concerning the effect of age, age was a predictor of the phonemic discrimination task in the SSD group, but not in the TD group. On the other hand, age was a predictor of the lexical and phonetic judgement task in the TD group, but not in the SSD group. The effects of age in this study appear to depend on tasks and groups. However, the effect of age in the study by Edwards et al. [11] did not predict performance on a minimal-pair discrimination task across the SSD and young TD groups. Receptive vocabulary was not a predictor of any speech perception tasks in this study, whereas it was a major predictor in Edwards et al.’s [11] study. Moreover, PCC was not a predictor in the current study, which is consistent with the studies by Waldman et al. [53] and Arayathaweekun [29] who both reported that auditory discrimination did not correlate with speech production.

Phonological Awareness

The present study included three tasks: initial phoneme identification, rime awareness, and lexical tone awareness. The initial phoneme identification task was adapted from Yampratoom et al.’s task [43]; however, the comparability of the results with the current study may be limited due to changes to some distractor words. The mean score in our findings was 4.53 in the SSD group and 4.70 in the TD group, while Yampratoom et al. [43] reported a mean score of 4.50 for TD children from a total score of 10. In the current study, children found it difficult to understand the concept of initial phonemes. Therefore, the tester repeated the first phoneme during instruction to help a child recognise it. This task is at the onset-rime awareness level, in which children can match word onsets but are not required to segment or blend individual phonemes. Anthony and Francis [8] suggested that exposure to different oral languages influences the development of PA. Onset awareness skills in Thai-speaking children tended to improve during the school-age period, approximately 6–8 years old [35].

A recent scoping review on PA in Thai-speaking children reported that previous studies have only addressed rime awareness in primary school children, not among preschool children [25]. The results of the present study on preschool children showed no significant difference in performance on the rime awareness task between the SSD and TD groups. The effect of age was a predictor of rime awareness in both groups, and receptive vocabulary was a predictor of this task in the SSD group. Interestingly, in this study, rime awareness was the only task in the SSD group in which age and receptive vocabulary significantly predicted performance. Our findings were similar to a previous study by Preston and Edwards [20], which found that age and receptive vocabulary predicted 33% of the variance in overall PA scores. They suggested that insufficient phonological representations in children with SSD are related to PA problems and low receptive vocabulary. PCC was not a predictor of any PA tasks in the present study. Likewise, Tambyraja et al. [54] reported that children with SSD who were identified as poor readers received lower scores of phonological processing, including PA. However, their speech production, evaluated by PCC, did not show significantly lower performance than that of those without reading problems.

Both groups showed high performance in the lexical tone awareness task in the present study, especially the TD group, who showed a potential ceiling effect. Age was also not a predictor of this task. Onsuwan et al. [26] suggested that Thai-speaking children can produce and identify all five lexical tones at around 2 to 3 years of age and achieve mastery by ages 5 or 7. This may indicate that some children in this study may already have achieved mastery of lexical tones. However, a study of Mandarin-speaking children with SSD reported weaker performance in consonant and lexical tone perception in those with atypical phonological processes, but not in those with developmental phonological processes [51]. Chien et al. [51] suggested that this may be because children with atypical phonological processes have less detailed phonological representations of lexical tones. As mentioned earlier, there were few children with atypical phonological processes in the current study.

Nevertheless, attention should not be limited to lexical tones in tonal languages, as other language-specific phonetic features may also play an important role in children’s performance, such as the three-way voicing contrast in Thai word-initial plosives [55]. Given the limited research on tonal languages, further investigation is needed to understand how phonetic characteristics influence speech perception and PA.

Vocabulary and PA

Vocabulary is a reliable language predictor of PA in both TD children and children with speech or language impairment [20]. However, in the present study, receptive vocabulary was only a predictor of the rime awareness task in the SSD group. When children have difficulties in retrieving the phonological information of words (which are used for forming phonological representation), they may also have problems with recognising and manipulating speech sounds, such as rime awareness [20]. Metsala [23] also proposed that vocabulary growth may expand knowledge of phonological features, thereby assisting PA performance.

Clinical Implications

Although children with SSD in this study did not show difficulties in speech perception or PA, some individual children with SSD potentially performed outside the typical range. Children with SSD who scored two standard deviations below the mean were considered to fall below the typical range: two children in the phonemic discrimination task, one child in the rime awareness task, and three children in the lexical tone awareness task (including one child who fell below the typical range in both phonemic discrimination and lexical tone awareness). Therefore, we recommend that SLTs consider assessing these skills for early detection of those children who may have potential difficulties, consistent with previous studies suggesting these difficulties can be found in children with SSD [2, 4]. All five tasks in this study can be developed to assess speech perception or PA skills in children with SSD in the Thai context. In addition to speech perception and PA assessments, the ICS [39] is useful for assessing a child’s intelligibility from a parent’s perspective. Applying these assessments would assist SLTs in planning a comprehensive intervention for children with SSD for whom PA difficulties are identified. Interventions should address not only on speech production in children with SSD, but also on speech perception and PA. For example, enhancing vocabulary skills may support PA abilities, such as rime awareness, in these children [23].

Limitations

The current study did not collect data on socio-economic status or parental education, which are known to influence children’s emergent literacy skills [43, 56]. According to the Thailand National Statistical Office [57], the reported average monthly household income for the combined area of Bangkok and three surrounding provinces, such as Samut Prakan, was THB 37,405 in the first 6 months of 2025. This area includes all locations from which participants were recruited in the current study. By comparison, the majority of households in Thailand (39%) earned THB 15,001 to THB 30,000 per month. This may suggest that the average household income in Bangkok and the surrounding areas is generally higher than the national average, although the mean could be influenced by higher income households.

Regarding recruitment, the SSD group was recruited from both hospitals and schools, whereas the TD group was recruited from schools only. Although this difference might raise concerns about potential selection bias, it should be noted that all children with SSD in the current study attended schools in the same geographical area. Moreover, the range of SSD severity in children recruited from hospitals was similar to that of children recruited from schools, suggesting that recruitment context did not systematically favour one group over another.

Recommendations for Future Research

Children with SSD included in this study presented with mild-to-moderate severity of SSD. They also presented with articulation and phonological errors, which may lead to varied performance. Therefore, recruiting participants with more severe SSD profiles and with the same subtype of SSD (articulatory or phonological) in future studies may provide more explicit evidence of difficulties in speech perception and PA tasks. Moreover, future research is needed to develop Thai standardised articulation or phonology assessments, as none are currently available, making it difficult to categorise children into specific SSD subtypes.

Future studies could include a larger sample and recruit a broader age range (i.e., potentially younger) to capture better the developmental and performance differences between the two groups. The present study found potential ceiling effects; consequently, future studies should refine the task design or focus on younger children, while maintaining task difficulty appropriate to participants’ abilities. Moreover, including participants from areas outside the Bangkok Metropolitan Region may reflect variations in children’s performance across different dialects and socio-economic statuses.

Conclusion

This study found that Thai-speaking children with SSD did not perform differently from their TD peers in speech perception and PA tasks when controlling for age and receptive vocabulary. Age was a predictor of performance on phonemic discrimination and rime awareness tasks in the SSD group, and on lexical and phonetic judgement and rime awareness tasks in the TD group. Receptive vocabulary was a predictor only of rime awareness task in the SSD group. Speech production, as evaluated by PCC, was not a significant predictor of any task in either group. Children with SSD did not exhibit difficulties in speech perception or PA, which may be attributable to the fact that the children in this study presented with only mild-to-moderate levels of SSD severity. Additionally, potential ceiling effects were observed in the current study, which may have limited the ability to detect group differences. Nevertheless, although children with SSD did not show difficulties with speech perception or PA, this does not necessarily mean they have equivalent abilities to TD children. Further research is needed to investigate subgroups of SSD and to better understand the underlying deficits.

Acknowledgment

We would like to thank Dr. Paranat Muangsuwan for his assistance in transcribing samples for the inter-rater assessment.

Statement of Ethics

This study protocol was reviewed and approved by the Department of Psychological Sciences and Health Ethics Committee, University of Strathclyde, UK, Approval No. 64.19.01.2024.A., and the Institutional Review Board, Faculty of Medicine Ramathibodi Hospital, Mahidol University, Thailand, Approval No. MURA2024/485. Written informed consent was obtained from parents or legal guardians for their children to participate in the study.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

This work was conducted as part of a PhD study sponsored by the Faculty of Medicine Ramathibodi Hospital, Mahidol University, Thailand.

Author Contributions

Ketnipa Ratanakul: conceptualisation, data curation, formal analysis, investigation, methodology, and writing – original draft, review, and editing; Joanne Cleland and Wendy Cohen: conceptualisation, methodology, supervision, and writing – review and editing; Dwight C.K. Tse: formal analysis and writing – review and editing.

Funding Statement

This work was conducted as part of a PhD study sponsored by the Faculty of Medicine Ramathibodi Hospital, Mahidol University, Thailand.

Data Availability Statement

The anonymised data from the experiment will be deposited in the university repository upon completion of the project. Further enquiries can be directed to the corresponding author.