Musical Advantage in Lexical Tone Perception Hinges on Musical Instrument: A Comparison between Pitched Musicians, Unpitched Musicians, and Nonmusicians

Although previous studies found causal and correlational evidence of music-to-language transfer in lexical tone perception (e.g., Burnham et al., 2015; Delogu et al., 2010; Nan et al., 2018), one important component is missing: does the type of musicianship matter for music-to-language transfer? In the OPERA framework, the Precision requirement is that music training must entail more precise processing of the common acoustic feature of the music and language perceptual attributes (Patel, 2011, 2014). Accordingly, music training should enhance lexical tone perception only if it has a higher demand on pitch than does speech. As described below, different musical instruments have different processing demands on pitch. However, existing lexical tone perception studies have not considered this factor when examining music-to-language transfer (e.g., Choi, 2020; Cooper & Wang, 2012; Zheng & Samuel, 2018). Most often, the musician group contained a heterogeneous sample of musicians who had learnt musical instruments with different pitch processing demands (e.g., piano and drum).

Unlike lexical tone perception studies, many music perception studies have considered the heterogeneity of musicianship such as genre (Tervaniemi et al., 2016) and the type of musical instrument (Cameron & Grahn, 2014; Cicchini et al., 2012). Concerning genre, the timing of the last sound in a melodic sequence has a larger expressive importance in classical music and jazz relative to rock (Tervaniemi et al., 2016). Compared with rock musicians, classical and jazz musicians showed a stronger mismatch negativity (MMN) response to timing delay in the end of melodic sequences (Tervaniemi et al., 2016). Regarding the type of musical instrument, percussion instruments (e.g., drums) have a higher demand on rhythm than non-percussion instruments (e.g., violin) (Vuust et al., 2012). As expected, percussionist musicians outperformed non-percussionist musicians on musical rhythmic perception (Cicchini et al., 2012; cf. Slater & Kraus, 2015) and reproduction (Cameron & Grahn, 2014).

Of interest in this study is pitch. Based on pitch processing demand, musical instruments can be broadly classified as pitched and unpitched musical instruments. Pitched musical instruments require players to control the pitch over an equal-tempered scale (e.g., piano) or continuously monitor and manipulate a constantly changing pitch (e.g., violin). Relative to nonmusicians, pianists and violinists showed larger auditory evoked potential to musical pitch (Shahin et al., 2003), reflecting enhanced musical pitch perception. Thus, we included pianists and violinists in the pitched musician group. Focusing on rhythm, unpitched musical instruments produce sounds of indefinite pitch and therefore have minimal pitch processing demand (e.g., bass drum and cymbal) (Alexander et al., 2005; Parker, 1983).

To extend the previous studies, we investigated whether pitched musicians exhibit a unique advantage in tone perception relative to unpitched musicians and nonmusicians. According to OPERA, enhanced subcortical pitch processing undergirds the music advantage in lexical tone perception. Since pitched musical instruments have a high pitch processing demand, we hypothesized that the pitched musicians would outperform the unpitched musicians and the nonmusicians. As unpitched musical instruments have a minimal pitch processing demand, we predicted that the unpitched musicians and nonmusicians would perform similarly. Unlike in the previous studies, our nuanced classification of pitched and unpitched musicians has enabled us to comprehensively test the Precision element of OPERA in the correlational context (e.g., Choi, 2020; Lee et al., 2014; Zheng & Samuel, 2018).

In short, we compared pitched musicians, unpitched musicians, and nonmusicians on lexical tone discrimination and sequence recall. According to meta-analyses, music training enhances cognitive skills such as working memory and nonverbal intelligence (Bigand & Tillmann, 2022; Talamini et al., 2017). Since AXB discrimination and especially sequence recall have a high cognitive demand, we statistically controlled for these factors (Chen et al., 2020). If the Precision element holds, the pitched musicians should outperform the unpitched musicians and the nonmusicians on lexical tone perception. Moreover, the unpitched musicians should perform similarly to the nonmusicians. To recap, the following research questions motivated our study:

1. Did pitched musicians outperform unpitched musicians and nonmusicians on lexical tone discrimination and sequence recall?

2. Did unpitched musicians perform similarly to nonmusicians on lexical tone discrimination and sequence recall?

We recruited 61 Cantonese listeners (33 males and 27 females) aged from 19 to 33 years (M = 22.8 years; SD = 2.61 years) in Hong Kong via email and posters. All participants provided informed consent and completed a language and music background questionnaire (Choi, 2021, 2022a; Choi & Lai, 2023). According to self-report, all Cantonese listeners spoke Cantonese as first language, had normal hearing, and had no previous Thai learning experience and absolute pitch. We classified the participants into three groups, i.e., pitched musicians, unpitched musicians, and nonmusicians. Based on the criteria used in previous studies (Choi, 2020, 2022b; Tong et al., 2018), the pitched musicians had at least seven years of continuous piano and/or violin training, less than two years of unpitched percussion training, and could play their instruments at the time of testing. The unpitched musicians had at least seven years of continuous unpitched percussion training, less than two years of pitched music training, and could play their instruments at the time of testing. The nonmusicians had less than two years of music training, no music training in the past five years, and could not play any musical instrument at the time of testing. We excluded three unpitched percussionists from the study due to excess pitched instrument learning experience. The final sample contained 20 pitched musicians (10 males, 10 females), 18 unpitched musicians (11 males, 7 females) and 20 nonmusicians (10 males, 10 females).

The mean ages of the pitched musicians, unpitched musicians and nonmusicians were 21.95 years (SD = 1.79 year), 23.72 years (SD = 3.17 years), and 23.00 years (SD = 1.97 year), respectively. On average, the pitched musicians and unpitched musicians had 10.35 years (SD = 3.42 years) and 9.50 years of music training (SD = 2.81 years), respectively, while the nonmusicians only had 1.30 years of music training (SD = 0.45 year) if at all. Tables 1 and 2 summarize the instruments learnt by the pitched and unpitched musicians.

The experiment took place in a sound booth at The University of Hong Kong or in a music studio. Participants completed the Thai tone discrimination and sequence recall tasks under the supervision of the second or the third author. We ran the experiment on a laptop with E-prime 3.0. Throughout the experiment, participants heard all audio stimuli via Sennheiser HD280 PRO headphones. The entire experiment lasted for about 40 minutes. We had obtained ethical approval from the Faculty Research Ethics Committee of the Faculty of Education at The University of Hong Kong.

Thai language has five lexical tones, including three level tones (high, mid, and low) and two contour tones (falling and rising), e.g., ปา [paa – mid (M)] throw, ป่า [pàa – low (L)] forest, ป้า [pâa – falling (F)] aunt, ป๊า [páa – high (H)] father, ป๋า [păa – rising (R)] dad (Cooke, 1963; Winskel, 2011). This yielded 10 Thai tone contrasts; that is, M-L, M-F, M-H, M-R, L-F, L-H, L-R, F-H, F-R, H-R.

Two native Thai speakers (one male and one female) recorded the stimuli with a Shure SM58 microphone at a sampling rate of 48k Hz. The two speakers naturally produced the five Thai tones embedded in different segments, e.g., /goht/, /glaa/, /yoo/, /miia/, and /meet/. To preserve the naturalness of the stimuli, we did not acoustically manipulate them (e.g., Choi, 2020; Cooper & Wang, 2012; Nan et al., 2018). The natural duration of the stimuli used in our study ranged from 420 ms to 1,394 ms. Although the decibels relative to full scale (RMS DB FS) was not equated between all the stimuli, our acoustic-behavioural analysis showed that the accuracies (of pitched musicians, unpitched musicians, nonmusicians, and all groups combined) were not associated with it, ps = .749, .587, .774, and .948. Before and during the experiment, participants could adjust the volume of the laptop. All participants reported that they could hear the stimuli clearly at a comfortable volume. The stimuli are available in Open Science Framework (https://osf.io/2p6hb).

We adopted the AXB paradigm (Choi, 2020). The task contained 10 pairs of Thai tone contrasts (i.e., M-L, M-F, M-H, M-R, L-F, L-H, L-R, F-H, F-R, and H-R). On each trial, listeners heard three syllables (e.g., /goht-M/ /goht-R/ /goht-R/) at 800 ms apart. Then, they pressed the associated keys to indicate if the first or the third syllable carried the same lexical tone as the second one did. AB and X were produced by speakers of different genders to prevent the participants’ reliance on simple acoustic comparisons. There were eight trials for each Thai tone contrast and the total number of trials was 80 (10 pairs of Thai tone contrasts x 8 trials). Prior to the task, participants were explicitly requested to attend to the lexical tones, but not duration, intensity, and voice. The task began with four practice trials with feedback. In each experimental trial, the accuracy was recorded and no feedback was provided. The sample-specific internal consistency was satisfactory (Cronbach’s α = .78).

The same native Thai speakers as above recorded a vowel contrast, i.e., /r∊-M/-/ru-M/, and two lexical tone contrasts, i.e., /phı-M/-/phı-H/ and /pluuk-F/-/pluuk-R/. The vowel contrast contained the same mid tone. The first tone contrast contained two level tones (mid and high) whereas the second one contained two contour tones (falling and rising). The stimuli are available in Open Science Framework (https://osf.io/2p6hb).

Adapted from the stress sequence recall task (Dupoux et al., 2010), this task assessed the ability to represent and store Thai tones in memory. The task contained the vowel context, the level tone context (/phı-M/-/phı-H/), and the contour tone context (pluuk-F/-/pluuk-R/). Each context began with a familiarization phase in which participants listened to two items (as many times as desired) and associated them with their corresponding keys [1] and [2]. Then, there were five identification trials with feedback. In the sequence recall phase, participants listened to a sequence of syllables (600 ms apart) with varying length (two to six) and reproduced the sequence by pressing the associated keys, e.g., pressing [1] [2] [2] [1] for /phı-M/-/phı-H/-/phı-H/-/phı-M/. On each trial, a response was considered correct only if it fully matched the sequence presented. In each context, there were six trials for each sequence length. The total number of trials was 60. The sample-specific internal consistencies were high in the vowel context (Cronbach’s α = .80) and very high in the level tone (Cronbach’s α = .90) and contour tone contexts (Cronbach’s α = .91).

We adapted the backward digit span task from the Wechsler Adult Intelligence Scale (Fourth Edition, Hong Kong; Wechsler, 2014). WAIS-IV(HK) was specifically developed for the Cantonese speaking population in Hong Kong aged between 16 and 90 years. The task was administered in Cantonese. On each trial, participants listened to a sequence of digits (two to eight) verbally produced by the experimenter in Cantonese, and recalled the sequence in reverse order. The sequence length increased by one after correct responses of two sequences at the same length until the participant failed to recall both sequences. The maximum sequence length recalled by each participant was collected. The sample-specific internal consistency was fair (Cronbach’s α = .65).

We conducted the Raven’s 2 Progressive Matrices to assess nonverbal intelligence (Raven et al., 2018). In each of the 24 questions, participants chose the appropriate picture to complete a visual pattern. One point was given for each correct answer. The total score was collected. The sample-specific internal consistency was satisfactory (Cronbach’s α = .73).

We conducted two sets of one-way ANOVAs to examine whether the three groups differed in working memory and nonverbal intelligence. The main effect of group was not significant for working memory, p = .402, and nonverbal intelligence, p = .095. These results indicate that the three groups matched on working memory and nonverbal intelligence.

To be empirically stringent, we still controlled for working memory and nonverbal intelligence in the main analysis below. The same set of main analysis without the two control variables yielded entirely consistent results. For both sets of analysis, we have uploaded the annotated.jasp files including data and input options to Open Science Framework (https://osf.io/2p6hb).

Our two research questions were: 1) whether pitched musicians outperformed unpitched musicians and nonmusicians on Thai tone discrimination and sequence recall; and 2) whether unpitched musicians performed similarly to nonmusicians in these tasks. As reasoned in  Appendix A, we will adopt the Bayesian approach to address these research questions.

We conducted a Bayesian two-way ANCOVA on mean accuracy with JASP 0.17.3 (JASP Team, 2023). The within-subject factor was lexical tone contrast (M-L, M-F, M-H, M-R, L-F, L-H, L-R, F-H, F-R, and H-R), the between-subjects factor was group (pitched musicians, unpitched musicians, and nonmusicians), and the covariates were working memory and nonverbal intelligence. In total, there were 20 models. We conducted model comparisons relative to the best-fit model (see Table 3). Bayes factors indicated that the data best supported the model (Tone + Group) with significant main effects of tone contrast and group. Specifically, BF₀₁ suggested that the null model and the other models were 3.96×10³⁷ and 1.67–1.10×10³⁹ times less likely than the best-fit model. Model comparisons relative to the null model yielded consistent results (see Table A2). Our central focus was the main effect of group. According to post hoc comparisons, the pitched musicians outperformed the unpitched musicians and the nonmusicians with moderate (BF_{10, U} = 4.47) and extreme evidence (BF_{10, U} = 189.46) respectively (see Table 4). However, the unpitched musicians and the nonmusicians performed similarly, with moderate evidence (BF_{10, U} = 0.19).

We conducted a Bayesian two-way ANCOVA on mean accuracy with context (vowel, level tone, and contour tone) as the within-subject factor, group (pitched musicians, unpitched musicians, and nonmusicians) as the between-subjects factor, and working memory and nonverbal intelligence as the co-variates. In total, there were 20 models. We conducted model comparisons relative to the best-fit model (see Table 5). Bayes factor indicated that the data best supported the model (Context + Group + IQ + Context × Group) with significant main effects of context, group, nonverbal intelligence, and the interaction between context and group. Specifically, BF₀₁ suggested that the null model and the other models were 5.40×10⁶ and 1.28–1.68×10⁷ times less likely than the best-fit model. Model comparisons relative to the null model yielded consistent results (see Table A3). Our central focus was the interaction between context and group.

To unpack the interaction between context and group, we conducted a Bayesian one-way ANCOVA in each context with mean accuracy as the dependent variable, group (pitched musicians, unpitched musicians, and nonmusicians) as the between-subjects factor, and working memory and nonverbal intelligence as the covariates. In the vowel context, the best-fit model was the model (Group + IQ) with significant main effects of group and nonverbal intelligence. Specifically, BF₀₁ indicated that the null model and the other models were 17.10 and 1.04–40.02 times less likely than the best-fit model (see Table 6; see also Table A4 for BF₁₀). For the main effect of group, post hoc comparisons indicated that the pitched musicians and unpitched musicians outperformed the non musicians with moderate evidence, BF_{10, U} = 7.68 and 7.92. However, the pitched musicians and unpitched musicians performed similarly with moderate evidence, BF_{10, U} = 0.326 (see Table 7).

In the level tone context, the best-fit model was the model (Group + IQ) with significant main effects of group and nonverbal intelligence. Specifically, BF₀₁ suggested that the null model and the other models were 3233.06 and 1.10–10666.53 times less likely than the best-fit model (see Table 8; see also Table A5 for BF₁₀). For the main effect of group, post hoc comparisons showed that the pitched musicians outperformed the unpitched musicians and the nonmusicians with moderate evidence (BF_{10, U} = 3.06) and extreme evidence (BF_{10, U} = 12109.23) respectively. Moreover, the unpitched musicians outperformed the nonmusicians with moderate evidence, BF_{10, U} = 4.75 (see Table 9).

In the contour tone context, the model comparisons favored the null model. Specifically, BF₀₁ suggested that the other models were 1.24–7.98 times less likely than the null model (see Table 10).

To supplement the above analysis, we also unpacked the interaction between context and group at each level of group. The results are summarized in  Appendix B.

The present study examined: 1) whether pitched musicians outperformed unpitched musicians and nonmusicians on Thai tone discrimination and sequence recall; and 2) whether unpitched musicians performed similarly to nonmusicians in these tasks. For lexical tone discrimination, the pitched musicians outdid the unpitched musicians and the nonmusicians. Moreover, the unpitched musicians performed similarly to the nonmusicians. Regarding lexical tone sequence recall, the pitched musicians outperformed the unpitched musicians and the nonmusicians in the level tone context. Moreover, the unpitched musicians also outdid the nonmusicians on level tone sequence recall. However, the three groups performed similarly in the contour tone context.

The principal finding is the unique advantage of pitched musicians in lexical tone discrimination. Consistent with our hypothesis, the pitched musicians discriminated Thai tones more accurately than the unpitched musicians and the nonmusicians. Importantly, the unpitched musicians did not outperform the nonmusicians. This has reflected that the musical advantage in lexical tone discrimination is unique to pitched musicians. Furthermore, the above results apply to all the Thai tone contrasts. On the one hand, our results offer correlational evidence for the Precision element of OPERA (Patel, 2011, 2014). For music-to-language transfer to occur in lexical tone perception, the Precision element assumes that music must entail more precise pitch processing than speech does. Pitched musical instruments have a high demand on pitch processing, so our pitched musicians exhibited an advantage in lexical tone discrimination. By contrast, unpitched musical instruments have minimal demand on pitch processing. Thus, our unpitched musicians showed no advantage in lexical tone discrimination relative to the nonmusicians. Without considering the heterogeneity of musical instruments, previous lexical tone perception studies suggested that music-to-language transfer occurred with music training (e.g., Choi, 2020; Lee et al., 2014; Zheng & Samuel, 2018). Extending these studies, our study further suggests that music-to-language transfer in lexical tone discrimination occurs only when the music training has a high processing demand on pitch, consistent with OPERA’s notion (Patel, 2011, 2014).

On the other hand, it is also possible that our pitched musicians had higher pre-existing musical pitch sensitivity than the unpitched musicians and the nonmusicians even well before music training (see Schellenberg, 2015). From an empirical perspective, we have provided correlational evidence to support the feasibility of an RCT study that compares the effectiveness of pitched versus unpitched instrumental training in lexical tone perception. If the pitched instrumental training group shows larger gains than the unpitched instrumental training and waitlist control groups, it would provide OPERA with strong causal evidence of the Precision element.

Consistent with our hypothesis, the pitched musicians outperformed the unpitched musicians and the nonmusicians on level tone sequence recall. Unexpectedly, even though the unpitched musicians underperformed the pitched musicians, the former still outdid the nonmusicians. Unlike the perceptual-based discrimination task, the sequence recall task entails both perception and sequence recall. Participants not only had to perceive the lexical tones, but also remember and reproduce the sequences. Music training typically requires musicians to remember and reproduce melodic and rhythmic sequences (Bigand & Tillmann, 2022; Talamini et al., 2017). This may translate to the cognitive ability to remember and reproduce sequences in the sequence recall task, benefiting the pitched and the unpitched musicians. On top of this cognitive enhancement on sequence recall which both groups enjoyed, the pitched musicians had better level tone discrimination. This further enabled the pitched musicians to encode the level tone sequences more efficiently than the unpitched musicians. Thus, the pitched musicians exhibited the largest musical advantage, and the unpitched musicians enjoyed a small musical advantage relative to the nonmusicians. Critically, our working memory measure (backward digit span) did not correlate significantly with level tone sequence recall (p = .578). Possibly, if we had controlled for a working memory measure that is more closely related to sequence recall (e.g., forward span), or even other cognitive measures (Benz et al., 2016; Zuk et al., 2014), the unpitched musicians might have performed similarly to the nonmusicians.

Inconsistent with our hypothesis, there was no musical advantage in contour tone sequence recall. At first glance, this finding paralleled a previous study on the selectivity of musical advantage in English musicians (Choi, 2020). However, there are two major discrepancies between the current results and the earlier work. First, English musicians outdid English nonmusicians on recalling contour but not level tone sequences, but Cantonese musicians showed the opposite pattern. This might be due to differences in their first language (English and Cantonese), the stimuli (Cantonese and Thai), and the focus of their music training. The most important discrepancy is that there are different underlying reasons for the observed selectivity of musical advantage between the two studies. In the previous study, English musicians performed worse on level than contour tone sequences, implying that musicianship did not benefit level tone sequence recall much, if at all (Choi, 2020; cf. Schellenberg, 2015). Contrastively, the Cantonese musicians recalled both sequences with similar accuracy, but the nonmusicians performed better on contour than level tone sequences. Unlike in the previous study, the loss of musical advantage was because the nonmusicians caught up with the musicians in recalling contour tone sequences. Although the nonmusicians discriminated contour tones less accurately than the pitched musicians in the AXB task, they might have developed a compensatory strategy to achieve the same performance as the pitched musicians in sequence recall. Future studies are needed to identify their compensatory strategy and why the strategy did not apply to level tones. From a theoretical perspective, the collective findings suggest that the Precision element may not apply to the recall of every lexical tone sequence. Moreover, there could be different reasons giving rise to the selectivity of musical advantage. For example, it could be musicianship not being associated with enhanced perception of some lexical tone contrasts (Choi, 2020), or nonmusicians catching up with musicians.

An incidental finding is that the pitched and unpitched musicians recalled vowel sequences more accurately than the nonmusicians. Acoustically, /∊/ and /u/ differ in formant frequencies and duration. Loui et al. (2011) proposed that musical pitch sensitivity underlies the processing of first and second formant frequencies. Moreover, Tierney and Kraus (2014) suggested that rhythmic sensitivity aids the perception of fine timing details in speech. Indeed, our pitched and unpitched musicians showed an advantage, reflecting the roles of pitch and rhythm in vocalic perception. Furthermore, the equal performance of the pitched and the unpitched musicians further suggests that pitch and rhythm may play equally important roles. We encourage researchers to conduct RCT studies on the effectiveness of pitch and rhythm training on vowel perception.

We have identified several avenues for future research. First, our correlational evidence suggests that it is feasible to conduct an RCT study on the causal effect of pitched and unpitched instrumental training on lexical tone perception. Second, our pitched musician group contained pianists and violinists. While pianists control the pitch over an equal-tempered scale, violinists need to monitor and regulate a constantly changing pitch (with finger placement, bow pressure, and string tension) (Shahin et al., 2003). Compared with level tones, contour tones have a dynamic F0 contour (Choi, 2020). Thus, researchers can examine whether violinists exhibit a larger musical advantage in contour tones compared with pianists. Third, musicianship is a multifaceted concept that is well beyond musical instrument. Cross-culturally, Western music pays more attention to steady-state pitch sequences (Leech-Wilkinson, 2006) whereas Chinese music has a high occurrence of gliding pitches (Shi, 2016). Researchers can examine the differential effects of Western and Chinese music training on level and contour tone perception. Lastly, researchers have started to distinguish between professional musicians and amateur musicians (Oechslin et al., 2013; Rogenmoser et al., 2017; Schneider et al., 2002) and enduring musicians and former musicians (Toh et al., 2023). These fine distinctions prompt nuanced investigations of music-to-language transfer.

To conclude, pitched musicians showed a unique musical advantage in lexical tone discrimination. In level tone sequence recall, although both pitched and unpitched musicians outdid the nonmusicians, the pitched musicians had the largest musical advantage. Taken together, the pitch processing demand of musical instrument may matter for music-to-language transfer in lexical tone discrimination and level tone sequence recall. From a theoretical perspective, this offers correlational support for the Precision element of OPERA (Patel, 2011, 2014). From a practical perspective, there is a trend of utilizing music training to enhance speech perception (e.g., Kraus & Chandrasekaran, 2010; Nan et al., 2018). Our study further suggests that the choice of musical instrument may matter, at least for lexical tone perception. In addition to musical instrument, there are many more ways to decipher musicianship (e.g., Oechslin et al., 2013; Rogenmoser et al., 2017; Toh et al., 2023). Deciphering musicianship can certainly advance the theoretical understanding of music-to-language transfer and its practical application.

We have no known conflict of interest to disclose. This study was based, in part, on the undergraduate final year projects of the second and third authors. We appreciate Kusol Im-Erbsin and Ratana-U-Bol for their assistance in stimuli development. This work was supported by The University of Hong Kong (Seed Fund for Basic Research for New Staff [202107185043] and Project-Based Research Funding [supported by Start-up Fund]).

Our second research question was whether unpitched musicians performed similarly to nonmusicians in lexical tone discrimination and sequence recall. Statistically, null hypothesis significance testing either rejects or does not reject the null hypothesis. Thus, it cannot differentiate between absence of evidence and evidence of absence (Ly et al., 2019). Simply put, the (potential) lack of significant difference between the unpitched musicians and the nonmusicians (i.e., p > .05) would not necessarily indicate that the two groups performed similarly. In that case, we could only conclude that we had found no evidence that the two groups differed. To overcome this issue, we adopted Bayesian hypothesis testing (Dienes, 2014, 2016).

Bayesian hypothesis testing can provide both evidence of absence and presence. Unlike null hypothesis significance testing, the Bayesian approach measures the strength of evidence for both the null and alternative hypotheses (Dienes, 2014, 2016). Bayes factor (BF₁₀ or BF₀₁) is the ratio of the likelihood of the data given one hypothesis to the likelihood of the data given another hypothesis. For example, a BF₁₀ of 30 indicates that the data are 30 times more likely under the alternative hypothesis than the null hypothesis. By contrast, a BF₀₁ of 30 suggests that the data are 30 times more likely under the null hypothesis than the alternative hypothesis. Table A1 summarizes the Bayes factors and their respective evidence strength (Lee & Wagenmakers, 2013). Due to the lack of previous findings, we used a uniform prior that is relatively objective (Jaynes, 2003).

To unpack the interaction between context and group, we conducted a Bayesian one-way ANCOVA in each group with mean accuracy as the dependent variable, context (vowel, level tone, and contour tone) as the within-subject factor, and working memory and nonverbal intelligence as the covariates. For the pitched musicians, the best-fit model was the null model. BF₀₁ indicated that the other models were 1.2–13.99 times less likely than the null model (see Table A6).

For the unpitched musicians, the best-fit model was the model (Context + IQ) with significant main effect of context and IQ. BF₀₁ revealed that the other models were 1.78–16.84 times less likely than the best-fit model (see Table A7). For the main effect of context, post hoc comparisons showed that the unpitched musicians performed significantly better in the vowel context than the level tone context and the contour tone context with moderate evidence (BF_{10, U} = 4.49) and anecdotal evidence (BF_{10, U} = 1.39), respectively. They performed similarly across the level tone and contour tone context with anecdotal evidence, BF_{10, U} = 0.36 (see Table A8).

For the nonmusicians, the best-fit model was the model (Context) with a significant main effect of context. BF₀₁ revealed that the null model and the other models were 473.49 and 2.21–2111.09 times less likely than the best-fit model (see Table A9). For the main effect of context, post hoc comparisons indicated that the nonmusicians performed significantly better in the contour tone and vowel contexts than in the level tone context, with very strong (BF_{10, U} = 64.41) and extreme evidence (BF_{10, U} = 414.27), respectively. However, they performed similarly in the contour tone and vowel contexts with moderate evidence, BF_{10, U} = 0.25 (see Table A10).