Cortical Tracking of Speech and Music Predicts Reading Ability in Adults

1 Cortical tracking of acoustic features is essential for the neural processing of continuous 2 stimuli such as speech and music. It has been shown to vary based on language and music skills 3 and may reflect fundamental auditory temporal processing mechanisms supporting literacy. In 4 the current study, we tested whether cortical tracking of speech and music predicted reading 5 ability in healthy young adults (N = 32) , evaluated through a lexical decision task . Participants 6 first completed an online session in which they performed a lexical decision task to assess their 7 reading skills . This was followed by an electroencephalography (EEG) session , in which 8 participants listened to a naturalistic short story and a music track. Using mutual information, 9 we showed that neural activity aligned to both speech and music across a wide range of 10 frequencies. Interestingly, c ortical tracking was stronger for speech at very low frequencies, 11 while it was stronger for music at higher frequencies. Critically, cortical tracking predicted 12 reaction times in the lexical decision task in a frequency-dependent manner: stronger delta-band 13 tracking (~1-3 Hz) for both speech and music was associated with faster reaction times, whereas 14 stronger alpha -band tracking (~12 Hz) for speech was associated with slower reaction times . 15 These findings remained significant even when controlling for stimulus type, age, musical 16 experience and reading enjoyment . These results suggest that cortical tracking of speech and 17 music reflect a domain-general temporal processing mechanism that is associated with reading 18 ability beyond stimulus-specific features. 19 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability

20 Speech and music are (quasi-) rhythmic signals formed of hierarchically organised 21 structural units that unfold over time (Lerdahl & Jackendoff, 1983; Miyagawa et al., 2013; Asano, 22 2022). For speech, these units include phonemes and syllables , while for music , they include 23 notes and motifs. Aside from their rhythmic structure , speech and music share other 24 characteristics: they are both auditory signals structured by acoustic cues , such as duration 25 (timing), frequency (pitch) and amplitude (loudness). 26 Changes in these acoustic cues are reflected by the modulation of the signal’s amplitude 27 over time (Patel, 2010), often quantified as the acoustic envelope (Peelle & Davis, 2012). Neural 28 activity aligns with these slow envelope fluctuations of continuous external stimuli , known as 29 cortical tracking (Giraud & Poeppel, 2012; Doelling & Poeppel, 2015; Ding et al., 2017). Whether 30 this alignment is due to synchronisation of endogenous oscillations to an external rhythm or to 31 evoked responses is still under debate (Haegens & Golumbic, 2018; Obleser & Kayser, 2019), see 32 Atanasova et al. (2026) for a comprehensive review . Some factors which influence cortical 33 tracking are attention (Symons et al., 2021) , language and musical proficiency (Doelling & 34 Poeppel, 2015) , stimulus familiarity (Kumagai et al., 2017; Keitel et al., 2025) , and prior 35 knowledge (Sohoglu et al., 2012). 36 Neural activity in the brain track s the envelopes of speech and music at multiple 37 timescales (Giraud & Poeppel, 2012; Doelling & Poeppel, 2015; Ding et al., 2017) . In speech, 38 different temporal frequencies have been associated with various linguistic units: syllables are 39 largely associated with theta (4-8Hz), and words and phrases with delta (0.5-4Hz) (Ahissar et al., 40 2001; Peelle & Davis, 2012; Gross et al., 2013; Keitel et al., 2018) . A right hemisphere bias for 41 processing slow (syllabic/prosodic) information has been found (Abrams et al., 2008), whereas 42 the left hemisphere preferentially extracts information from faster temporal features (Obleser et 43 al., 2008) , consistent with the asymmetric sampling hypothesis (Poeppel, 2003) . Cortical 44 tracking has been linked to language and literacy development, though the precise nature of this 45 relationship remains complex. Research with dyslexic populations has yielded varied findings, 46 with some studies reporting reduced tracking of low -frequency acoustic information 47 (corresponding to syllabic and prosodic rhythms) and others demonstrating enhanced tracking 48 at higher frequencies (Lehongre et al., 2013; Molinaro et al., 2016; Power et al., 2016) . These 49 frequency-specific results may have direct implications for reading development, as r eading 50 relies upon cognitive mechanisms that parallel those underlying speech perception. Specifically, 51 graphemes (clusters of letters) , must be mapped onto phonological representations to enable 52 fluent decoding (Goswami, 2007). The temporal sampling framework (Goswami, 2011) proposes 53 that reading difficulties in dyslexia stem from impaired cortical tracking of speech at multiple 54 temporal scales. 55 Music, like speech, contains hierarchical temporal structures such as rhythm and meter, 56 which are also cortically tracked (Doelling & Poeppel, 2015; Harding et al., 2019; Keitel et al., 57 2025). These cortical tracking mechanisms are thought to overlap with those observed in speech 58 processing, particularly those used for parsing syllables and prosodic patterns, suggesting a 59 shared temporal processing system (Harding et al., 2019) , although it is currently still under 60

to what extent speech and music processing are domain -specific or domain -61 general(Kotz et al., 2018; Drakoulaki et al., 2024). 62 Due to rhythm deficits often found in dyslexic populations, researchers have investigated 63 the role of music (experience and training) related to language and reading, with many studies 64 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability finding that musical training leads to improved reading skills in children (Tierney & Kraus, 2013; 65 Garcia-de-Soria et al., 2025) . In particular, rhythm reproduction has been found to be strongly 66 associated with phonological awareness (Bhide et al., 2013; Goswami et al., 2013; Flaugnacco 67 et al., 2014) , however, causality remains contested. A recent review proposed that observed 68 differences in perceptual and cognitive abilities between musicians and non -musicians reflect 69 pre-existing aptitude differences (leading to selection bias) rather than effects of far transfer 70 (Schellenberg & Lima, 2024). 71 Here, we investigated whether cortical tracking of continuous natural speech and music, 72 quantified through mutual information, is associated with reading skills, quantified through a 73 lexical decision task , while controlling for age, musical sophistication and reading enjoyment . 74 We expected that stronger cortical tracking of the speech and music envelope s would predict 75 better performance in the lexical decision task (see preregistration: https://osf.io/t4dje/). 76 77

78 Participants 79 Thirty-two volunteers participated in this study (17 female, 19-28 years old; M = 22.2, SD 80 = 2.50). Participants declared never having received a diagnosis of neurological/psychological 81 disorders or dyslexia. All participants were right-handed (Oldfield, 1971). Self-reports of hearing 82 ability (Five-minute Hearing test, revised version) (Koike et al., 1994) indicated that 31 83 participants had no hearing impairments, while one participant was recommended a hearing test 84 (score of 23/60, with 20 being the cutoff) . This study was approved by the School of Social 85 Sciences Research Ethics Committee at the University of Dundee (approval number: UoD-SoSS-86 PSY-UG-2021-263) and adhered to the guidelines for the treatment of human participants in the 87 Declaration of Helsinki . Volunteers received monetary compensation of £10/h. Research 88 questions, hypotheses, measured variables and analyses were pre -registered on the OSF 89 website (https://osf.io/xrq36). Deviations from the pre-registration are detailed where necessary. 90 Procedure 91 Online procedure 92 Prior to taking part in the EEG experiment, p articipants were asked to complete the first 93 part of the study online, using the experiment builder Gorilla (Anwyl-Irvine et al., 2020) . In this 94 online session, participants completed a demographics questionnaire, a musicality assessment 95 (Müllensiefen et al., 2013), a handedness questionnaire (Oldfield, 1971) and the lexical decision 96 task. The musicality questionnaire included 14 items from the Goldsmiths Musical 97 Sophistication Index (GMSI) covering perceptual abilities, active engagement, singing abilit ies 98 and musical training. Each item included a statement such as “I spend a lot of my free time doing 99 music-related activities ”, which participants rated on a scale of 1 to 7 . The online task and 100 questionnaires had to be completed on a PC and in a single, uninterrupted session. 101 In-person procedure 102 After completing the online portion of the study, participants were invited to take part in 103 the in-person EEG session. Participants performed the EEG experiment in a soundproof booth 104 (160 cm x 110 cm) where they were seated approximately 65cm away from a 24 -inch monitor. 105 Participants were equipped with high-quality, wired headphones (Sennheiser, HD 25, 75Ω). The 106 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability experiment was run using Psychtoolbox (version ‘3.0.17)(Brainard & Vision, 1997; Pelli & Vision, 107 1997) through MATLAB (MATLAB, 2021). During the presentation of speech and music stimuli, 108 participants were instructed to relax, limit motor movement to avoid movement artefacts and 109 fixate on a small green circle on the screen. Blocks were self-paced, and the order of the auditory 110 stimuli was counterbalanced. 111 Stimuli and tasks 112 Lexical Decision Task 113 The lexical decision task was adapted from a pre -existing task (no reference was found 114 for the task or the study in which it was used) on the experiment-building website Gorilla (Anwyl-115 Irvine et al., 2020). The words and non-words used in this task were taken from a previous study 116 (Yao et al., 2018) and are part of the corpus of Glasgow Norms (Scott et al., 2019) . Both words 117 and nonwords ranged from 3 to 11 letters in length (M = 6.11, SD = 1.75). All words were concrete 118 nouns with neutral valence, as words with differing emotional connotations can affect response 119 times (Yao et al., 2018) . Word frequency (occurrences per million as per the British National 120 Corpus) was on average M = 23.95 (SD = 36.49). In the task, participants were asked to use the 121 left and right arrow keys to indicate whether the string of letters on the screen was a word (e.g., 122 ‘statue’) or a non-word (e.g., ‘depane’, which does not have any meaning in the English language). 123 Participants were given 12 practice trials with feedback. At the beginning of each block, there 124 were two brief screens, appearing for 1000 ms each, saying “Ready?” and “GO!” to prepare the 125 participant for the trials. A fixation cross was also shown for 500 milliseconds before each word 126 and nonword in the task. The experimental task consisted of 90 trials, given in 2 blocks of 45 trials 127 each, and feedback was not provided. All tasks and questionnaires used Open Sans font. 128 Halfway through the lexical decision task, participants were given the option to take a break and 129 recommence when they preferred. 130 Passive Listening to speech and music 131 All auditory stimuli were presented at a sampling rate of 44,100 Hz. The short story 132 selected for this study was “The Elves and the Shoemaker”, originally written by the brothers 133 Grimm, read by a female speaker with a pleasant voice (https://librivox.org/). The length of the 134 story was 300 s ( 5 minutes ). The articulation rate of the stimulus was calculated using Praat 135 (Boersma, 2001), through the automatic detection of syllable nuclei based on intensity peaks in 136 the speech signal (De Jong & Wempe, 2009) . The stimulus had an articulation rate of 3.61 137 syllables per second. Similarly, the modulation spectrum of the speech stimulus showed a peak 138 at 3.75 Hz (Figure 1A). 139 The music piece selected was “Fluid” by Lin Rountree, a pleasant jazz piece at 95 BPM. 140 This piece featured bass, keyboard and trumpet; there were no vocals included. The duration was 141 255 s (4 minutes and 25 seconds), and the modulation spectrum showed a peak at 3.15Hz (Figure 142 1A). After each piece, participants rated how familiar and how pleasant they had found the 143 stimuli by using a Visual Analog Scale: this involved placing a vertical marker between “did not 144 enjoy at all” and “enjoyed a lot” for enjoyment and “not at all familiar” and “extremely familiar” 145 for familiarity (each analysed in arbitrary units between 0 and 100). Participants rated the story 146 (M = 45.67, SD = 35.17) as significantly more familiar than the music piece (M = 27.27, SD = 25.17) 147 (t(31) = 2.55, p = .015, Figure 1C). However, enjoyment ratings did not differ significantly between 148 the story (M = 58.72, SD = 24.38) and the music piece ( M = 64.80, SD = 18.98) (t(31) = -1.08, p = 149 .287, Figure 1B). All stimuli are available on the OSF server (https://osf.io/xrq36). 150 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability 151 Figure 1 . (A) Modulation spectrum of speech (green) and music (purple) stimuli. Thick lines indicate 152 average values across 6 -s segments, and shaded areas represent the standard error of the mean. Peak 153 frequency is shown with dotted lines. (B) Enjoyment ratings for speech and music excerpts . There was no 154 significant difference between stimulus ratings (p = .287 ). (C) Familiarity ratings for speech and music 155 excerpts. The story was rated as more familiar than the music piece (t(31) = 2.55, p = .015). Note: Dots in 156 (B) and (C) show individual data points, violin plots show kernel density estimates , and boxplots show 157 median interquartile ranges and minimum/maximum. 158 Analysis of behavioural data 159 For analysis of the lexical decision task , incorrect trials (M = 5.69, SD = 3.25) were 160 excluded from analysis. For correct trials, responses faster than 250 ms or exceeding 1500 ms 161 were excluded (M = 6.58, SD = 10.45), as these could reflect inadvertent movements or lapses in 162 attention (Yao et al., 2018). Three participants were excluded from further analyses involving the 163 lexical decision task as their total number of rejected trials exceeded 2 SDs from the group mean. 164 For the remaining participants, median reaction times and accuracy rates were calculated 165 separately for word and nonword trials. These values were then averaged to yield total accuracy 166 and reaction time measures. 167 Acoustic envelope pre-processing 168 To analyse the neural tracking of speech and music signals, the wideband envelope of 169 the stimuli was extracted. The acoustic waveforms were filtered into eight frequency bands 170 (between 100 and 8000 Hz, 3rd order Butterworth filter, forward and reverse) equally distant on 171 the cochlear frequency map (Smith et al., 2002). The signal in each of these frequency bands was 172 then Hilbert -transformed and the magnitude extracted before being averaged to obtain the 173 wideband music and speech envelopes used in further analyses. Lastly, envelopes were down-174 sampled to a sampling rate of 150 Hz (Keitel et al., 2018). 175 EEG acquisition and pre-processing 176 EEG was recorded from 64 scalp electrodes and digitally sampled at 512 Hz , using a 177 BioSemi ActiveTwo system. Scalp electrodes were positioned according to the international 10-178 20 system. Electrodes with an offset of greater/less than ±20 mV were adjusted prior to starting 179 the recording. Ultimately, electrode offset for all electrodes was below an absolute value of 30 180 mV before the experiment began. Lateral eye movements were monitored by two electro -181 oculographic electrodes placed at the outer canthus of each eye. Vertical eye movements and 182 blinks were monitored by two electro -oculographic electrodes positioned below and above the 183 left eye. 184 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Data pre-processing was conducted using the Fieldtrip Toolbox (Oostenveld et al., 2011) 185 and custom -made scripts in MATLAB (R2025a) (The Mathworks, 2025) . The data were cut 186 according to the length of the stimul i, with an additional 2 -s leading and trailing window . Data 187 was first re-referenced to Cz, then, fourth-order Butterworth low-pass (60 Hz) and high-pass (0.2 188 Hz) filters were applied. Noisy (e.g., with higher variance than most) channels were visually 189 identified and interpolated through triangulation. A maximum of five channels was interpolated 190 per participant (M = 1.25, SD = 1.34). To remove eye artifacts and blinks, an independent 191 component analysis (ICA) was then carried out for 30 principal components . Components, 192 including eye movements or blinks, were selected and removed (M = 1.40, SD = 0.49). The EEG 193 signal was then down-sampled to 150 Hz to match the envelope signals (Keitel et al., 2018). 194 MI analysis 195 The correspondence between the continuous EEG signal and t he acoustic envelope 196 signals was analysed using a Gaussian copula mutual information framework (Ince et al., 2017). 197 The Mutual Information (MI) between the continuous L1-normalised EEG signal s and the 198 envelopes of the auditory stimuli (as well as their derivat ives, see Fig. 2 ) was calculated in the 199 frequency domain by applying a continuous wavelet transform (Chalas et al., 2022) , for 63 200 logarithmically spaced frequencies between .25 and 20 Hz. We used a participant-specific 201 optimal brain-stimulus lag computed by identifying, for each participant and condition, the lag at 202 which the MI value was highest (identified at electrode Cz for slow frequencies, averaged 203 between 0.5 – 4 Hz) . Each MI value was computed per participant, condition, frequency and 204 channel. To normalise MI, we first created random (surrogate) MI distributions. For this, we 205 segmented the continuous envelope signals into 5 -s segments and shuffled the segments 206 randomly. This kept the statistical properties of the signal but destroyed the temporal 207 relationship between the acoustic stimuli and brain signals. MI was then computed between the 208 brain signal and the shuffled envelope signals. Normalised MI was computed by z -scoring the 209 observed MI values against a surrogate distribution, subtracting the mean and dividing by the 210 standard deviation of randomised MI estimates. The normalised MI values were used for all 211 further analyses. 212 213 Figure 2. Brief excerpt (10 s) of acoustic envelope , waveform, and first derivative for the speech (green) 214 and music stimuli (purple). 215 216 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Statistical analysis 217 For each participant, normalised MI values at each channel and frequency were tested 218 against a null baseline of zero using dependent -samples t -tests. Cluster -based multiple -219 comparison correction (implemented in FieldTrip (Oostenveld et al., 2011) ) was applied using 220 5000 Monte Carlo permutations, in which the condition labels ( normalised MI vs. zero) were 221 randomly exchanged within subjects to form the null distribution. Clusters were defined as 222 spanning more than 3 frequency bins , and the cluster -level statistic was the sum of t-values. 223 Observed clusters were considered significant if their cluster statistic exceeded the 95th 224 percentile of the permutation distribution . This controlled for the family -wise error rate at the 225 cluster level. To avoid reporting spurious results, we only report clusters that cover a range of 226 frequencies of more than 0.5 Hz. 227 For the comparison between the two tracking conditions, dependent samples t-tests 228 were computed across channels and frequencies with MI values of one condition compared 229 against MI values of the other , with the same cluster -permutation approach used for multiple 230 comparison as described above. 231 To test the relationship between cortical tracking and behavioural measures (reaction 232 times from the lexical decision task) , Pearson’s correlations were computed between the 233 behavioural measures and the normalised MI values , across all electrodes and frequencies . 234 Before comparing the r values with the permutation distribution using cluster-based permutation 235 (using a minimum cluster size of 3 channels and tested against the 95 th percentile of the 236 permutation distribution), Pearson’s r values were transformed to be normally distributed using 237 Fisher’s z-transformation (Gorsuch & Lehmann, 2010). As an indicator of effect sizes, we report 238 Cohen’s d for peak electrodes. 239 To further compare the contribution of cortical tracking in both speech and music 240 conditions on performance in the lexical decision task , a robust multiple linear regression was 241 computed (R 4.5.1). The model included: (i) MI values, (ii) stimulus type (speech or music), (iii) 242 frequency band (delta or alpha), (iv) musical sophistication scores, (v) age , and (vi) reading 243 enjoyment as predictors , along with the three -way interaction between MI × stimulus type × 244 musical sophistication and the two -way interaction between MI × frequency band. Reaction 245 times in the lexical decision task served as the outcome variable . All continuous variables were 246 z-scored. 247 To explicitly test for hemispheric lateralisation in our significant correlation clusters, we 248 computed a lateralisation index (LI) following the procedure outlined by Haegens et al. (2011) 249 which is comparable to the method used by Thut et al. (2006). The LI was calculated as: 250 LI = (ipsilateral ROI – contralateral ROI) / (ipsilateral ROI + contralateral ROI) 251 where the region of interest (ROI) comprised the electrodes within each significant lateral 252 cluster. For each participant, correlation values ( r) from electrodes in one hemisphere were 253 paired with their hemispheric counterparts. The resulting LI values were then tested against zero 254 using a two -sided one -sample t-test to determine whether correlations differed significantly 255 between hemispheres. 256 257 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability

258 Behavioural results 259 In the word recognition task, a paired t-test found that reaction times to words (M = 260 658.53 ms, SD = 75.68) were significantly faster than to nonwords (M = 777.13 ms, SD = 106.87), 261 t(28) = -8.68, p < .001 (see Figure 3A). Accuracy was high for both words (M = 94.38%, SD = 3.83) 262 and nonwords (M = 92.87%, SD = 6.56) with no significant difference between conditions, t(28) = 263 1.04, p = 0.304 (Figure 3B) . The preregistration included both accuracy and reaction times as 264 indicators of reading ability. However, given that accuracy was at ceiling level and there was no 265 significant difference between conditions, we used only reaction times for the following 266 analyses. 267 268 Figure 3. Violin plot s and boxplots showing median reaction times to words and nonwords (A) and 269 accuracy for words and nonwords (B). Points indicate individual data for all participants; violin plots show 270 kernel density estimates and boxplots show median interquartile ranges and minimum /maximum. 271 Cortical tracking of the speech and music envelopes 272 We analysed whether participants tracked the acoustic speech and music envelopes 273 using dependent-samples t-tests of the normalised MI values (across channels and 274 frequencies) against a baseline of zero, with cluster-based permutation to control for multiple 275 comparisons. For speech (Figure 4A), we found a large positive cluster of all 64 electrodes, 276 ranging from 0.26 Hz to 12.37 Hz, which significantly tracked amplitude fluctuations with a peak 277 at 0.67 Hz (Cohen’s dpeak = 2.59, pcluster < .001). Similarly, for music (Figure 4B), we found a large 278 positive cluster of all 64 electrodes that significantly tracked amplitude fluctuations from 0.26 279 Hz to 18.75Hz with a peak at 5.38Hz (Cohen’s dpeak = 3.43, pcluster < .001). This indicates 280 widespread spatial and spectral tracking of both speech and music stimuli. 281 282 283 284 285 286 287 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability 288 289 Figure 4. Cortical tracking of acoustic speech (A) and music (B) envelopes. Channel-by-frequency 290 heatmaps of t-values indicating cortical tracking, with significant cluster s outlined in black. Below each 291 heatmap is the topography of the average cortical tracking across all significant frequencies with 292 significant channels highlighted in white. 293 Comparison of speech and music tracking 294 We then directly compared cortical tracking in speech and music conditions across 295 channels and frequencies using dependent -samples t-tests with cluster-based permutation to 296 control for multiple comparisons . Three significant clusters emerged: one negative cluster and 297 two positive clusters (Figure 5). A large negative cluster of 56 electrodes (Figure 5B(i)), ranging 298 from 0.26 Hz to 1.09 Hz (Cohen’s dpeak = 1.87, pcluster < .001) indicated that the speech envelope 299 was tracked more strongly than the music envelope at low frequencies. The two positive clusters 300 indicate that the music envelope was tracked more strongly than the speech envelope. The first 301 positive cluster (Figure 5B(ii)) included 18 temporal and occipital electrodes and ranged from 302 2.34 Hz to 3.55 Hz (Cohen’s dpeak = 1.48, pcluster = 0.032) while the second positive cluster (Figure 303 5B(iii)) included 59 electrodes and ranged from 1.66 Hz to 16.32 Hz (Cohen’s dpeak = 2.82, pcluster < 304 .001). 305 306 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability 307 Figure 5. (A) Mean t-values for the comparison between cortical tracking of speech and music across all 308 frequencies. Negative values indicate stronger tracking of speech (at low frequencies between 0.26 and 309 1.09 Hz). Positive values indicate stronger tracking for music (at frequencies between 1.66 and 16.32 Hz). 310 Shaded areas indicate the standard error of the mean. (B) Channel-by-frequency heatmap of t-values of 311 the comparison between cortical tracking of speech and cortical tracking of music. The negative cluster 312 (i), showing stronger cortical tracking of speech, is shown in orange. Positive clusters, showing stronger 313 tracking of music, are shown in pink ( ii) and green (iii). Topographies show the average cortical tracking 314 across all significant frequencies, with significant channels shown in white. 315 Cortical tracking predicts performance in the lexical decision task 316 To test whether envelope tracking during listening to the story and music predicted 317 participants’ behavioural performance in the lexical decision task, we correlated the MI values 318 per electrode and frequency with participants’ average reaction time across words and 319 nonwords. Note that separate initial analyses for reaction times for words and nonwords 320 indicated that results were comparable and we opted to use average reaction times to streamline 321 the presentation of results. 322 For the story condition, w e found one negative occipital cluster (Figure 6A(i)) at low 323 frequencies (0.89 Hz - 1.78 Hz), that significantly predicted reaction times in the lexical decision 324 task (Cohen’s dpeak = 1.39, pcluster < .001, 9 electrodes), which was right lateralised t(11) = -3.09, p 325 = .010 , indicating that participants who showed stronger envelope tracking to speech at low 326 frequencies had faster reaction times in the lexical decision task. 327 We also found one positive cluster (Figure 6A(i i)) in occipital and parietal electrodes at 328 high frequencies (10.05 Hz – 13.26 Hz) that predicted reaction times in the lexical decision task 329 (Cohen’s dpeak = 2.41, pcluster < .001, 6 electrodes), which was left lateralised ( t(25) = 9.155, p < 330 .001). This indicates that participants who showed stronger envelope tracking to speech at those 331 higher frequencies had slower reaction times in the lexical decision task. 332 For the music condition, w e found one negative cluster in frontal electrodes at low 333 frequencies (0.89 Hz – 1.44 Hz) that predicted reaction times in the lexical decision task (Cohen’s 334 dpeak = 2.40, pcluster < .001, 13 electrodes; Figure 6B). This indicates that participants who showed 335 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability stronger envelope tracking to music at low frequencies had faster reaction times in the lexical 336 decision task. 337 338 Figure 6. Relationship between cortical tracking and performance on the lexical decision task, quantified 339 through reaction times. (A) Channel-by-frequency heatmap of the correlation between cortical tracking of 340 speech and participants’ response speed i n the lexical decision task (reaction times). Two clusters were 341 found for speech : each topography shows the average r-values across frequencies of the cluster. The 342 negative cluster (i) indicated that stronger tracking of speech correlated with faster reaction times in the 343 lexical decision task. The positive cluster in the left hemisphere in occipital electrodes at high frequencies 344 (ii) indicated that stronger tracking of speech correlated with slower reaction times in the lexical decision 345 task. (B) Channel-by-frequency heatmap of the correlation between cortical tracking of music and 346 participants' response speed in the lexical decision task. The topography below shows the average r-347 values across frequencies of the negative cluster. Here, stronger tracking of music was associated with 348 better performance (i.e., faster reaction times). Note: Significant channels are highlighted with white 349 circles. 350 To assess the joint contribution of cortical tracking in both speech and music conditions 351 to lexical decision performance, and control for additional variables such as musical 352 sophistication, reading enjoyment and age , we performed an additional multiple linear 353 regression in R (Version 4.5.1). The model included: (i) MI values averaged within 354 positive/negative significant clusters, (ii) stimulus type (speech or music), (iii) frequency band 355 (delta or alpha), (iv) musical sophistication scores, (v) age , and (vi) reading enjoyment as 356 predictors, along with the three -way interaction between MI × stimulus type × musical 357 sophistication and the two-way interaction between MI × frequency. Reaction times in the lexical 358 decision task served as the outcome variable (Table 1). Importantly, if speech and music tracking 359 predicted reaction times differentially, we would expect a significant interaction between MI × 360 stimulus type. The main effect of cortical tracking (Figure 7A) was statistically significant ( β = -361 0.46, 95% CI [-0.75, -0.17], p = .003), indicating that, overall, participants with stronger cortical 362 tracking exhibited faster reaction times in the lexical decision task. This confirms the results of 363 the main analysis (Figure 6), while controlling for participant-related variables. Neither the main 364 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability effect of stimulus type ( p = .687) nor musical sophistication (p = .299) reached significance, 365 suggesting that these factors did not directly influence reaction times. 366 Figure 7. Relationship between cortical tracking and performance on the lexical decision task. (A) Main 367 effect of MI on reaction times, with the purple line indicating the prediction model and the dots showing 368 individual data points. (B) Main effect of age on reaction times, with the green line indicating the prediction 369 model and the dots showing individual data points. (C) Main effect of frequency band (alpha vs delta) on 370 reaction time. Dots indicate estimated marginal means from the regression model, and whiskers show 371 95% confidence intervals. (D) Two-way interaction: participants with higher MI in the delta frequency band 372 also show faster reaction times, where as participants with higher MI in the alpha frequency band show 373 slower reaction times in the lexical decision task. 374 Age showed a significant main effect (Figure 7 B; β = 0.20, 95% CI [0.02, 0.39], p = .034), 375 with older participants showing slower reaction times. Frequency also showed a significant main 376 effect (Figure 7C; β = 0.62, 95% CI [0.06, 1.18], p = .031) with cortical tracking in the alpha band 377 showing slower reaction times than in the delta band. The interaction between cortical tracking 378 and frequency (Figure 7 D) was statistically significant ( β = 1.67, 95% CI [ 1.02, 2.31], p <.001), 379 indicating that the relationship between MI and reaction times was dependent on the frequency 380 band. Specifically, stronger cortical tracking in the delta band was associated with faster 381 reaction times, whereas stronger tracking in the alpha band was associated with slower reaction 382 times, as shown in the main analysis (Figure 6) while controlling for musical sophistication, age 383 and reading enjoyment. No other two-way or three-way interactions were statistically significant 384 (all p-values > .280) 385 Table 1 386 Summary table for the robust linear regression model predicting reaction times 387 Predictors Estimates SE 95% CI p LL UL Intercept .08 .16 -.25 .41 .614 MI -.46 .14 -.75 -.17 .003 ** Stimulus [Music] -.10 .24 -.56 .37 .687 Musical Sophistication -.13 .12 -.36 .11 .299 Frequency [alpha] .62 .28 .06 1.18 .031 * Age .20 .10 .02 .39 .034 * Reading Enjoyment .05 .32 -.14 .24 .623 MI x Stimulus [music] .11 .23 -.34 .56 .616 MI x Musical Sophistication .01 .13 -.24 .26 .910 Stimulus [music] x Musical Sophistication -.15 .21 -.58 .27 .473 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability MI x Frequency [alpha] 1.67 .32 1.02 2.32 <.001 *** MI x Stimulus [music] x Musical Sophistication .23 .21 -.19 .65 .280 Note. Significant predictors are highlighted in bold. *p < .05, **p <.01, ***p < .001 388

389 The present study examined the relationship between cortical tracking of continuous 390 naturalistic speech and music and reading skills, quantified through a lexical decision task . 391 Several key findings emerged. First, cortical tracking predicted reading performance in a 392 frequency-dependent manner: stronger delta-band tracking was associated with faster reaction 393 times, whereas stronger alpha-band tracking was associated with slower reaction times. These 394 findings remained even when controlling for stimulus type, musical sophistication , age and 395 reading enjoyment. Secondly, both speech and music were found to be extensively tracked 396 across a wide range of channels and frequencies. Finally, cortical tracking of speech was found 397 to be stronger than that of music at low frequencies (<1 Hz), while cortical tracking of music was 398 found to be stronger at higher frequencies (1-16Hz). 399 Domain-general cortical tracking predicts reading performance 400 Cortical tracking of speech has been identified as a potentially critical mechanism 401 underlying language and literacy development, as suggested by studies demonstrating atypical 402 tracking patterns in children with dyslexia (Power et al., 2016; Araújo et al., 2024). We here tested 403 whether cortical tracking of speech , as well as music, would predict individual differences in 404 reading skill in healthy adults . Consistent with this hypothesis, cortical tracking emerged as a 405 significant predictor of reading performance. 406 There was no main effect of stimulus type on lexical decision task performance , 407 suggesting that both speech and music tracking contributed similarly to performance on the 408 lexical decision task. Although the absence of proof is no proof of absence, taken together, our 409 findings suggest that domain -general auditory processing mechanisms, rather than stimulus-410 specific features, underlie the relationship between cortical tracking and reading proficiency. 411 Notably, the topographies and frequency ranges (Figure 6A(i) and Figure 6B) look remarkably 412 similar, although for speech the effect reaches significance for posterior channels, while it 413 reaches significance for music in anterior channels. The absence of a stimulus effect aligns with 414 theoretical accounts proposing that shared temporal processing mechanisms support both 415 linguistic and musical perception (Patel, 2011) , and suggests that auditory processing 416 mechanisms contribute to reading proficiency (Cason et al., 2015) . We used a lexical decision 417 task as a proxy for readings skill, as this has been shown to be strongly associated with reading 418 proficiency (Weems & Zaidel, 2004; Lubineau et al., 2024) . However, reading is a multi -faceted 419 process, and future research should test whether the relationship between cortical tracking and 420 reading skill also applies to other tasks, such as reading comprehension, reading fluency, and 421 phonological decoding. 422 The relationship between cortical tracking and reading performance was modulated by 423 frequency band. Stronger tracking in the delta band predicted faster reaction times, whereas 424 stronger tracking in the alpha band predicted slower reaction times. This frequency -dependent 425 pattern suggests that successful reading may depend on the integrated processing of multiple 426 temporal scales. Enhanced delta -band tracking may facilitate the extraction of syllabic and 427 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability prosodic information critical for phonological processing (Goswami, 2011) while excessive 428 alpha-band tracking may reflect inefficient allocation of neural resources to fine -grained 429 temporal details at the expense of syllabic -level information (Peelle & Davis, 2012) . This 430 interpretation is consistent with the temporal sampling framework, which posits that reading 431 difficulties arise from impaired multi -scale temporal processing of speech (Power et al., 2016). 432 The observed association between stronger alpha -band tracking and slower reaction times 433 aligns with emerging evidence linking stronger high-frequency cortical tracking to reading 434 difficulties. Individuals with dyslexia and developmental language disorder exhibit atypical 435 frequency-specific tracking patterns (Araújo et al., 2024; Nora et al., 2024). Findings vary across 436 studies: some report only reduced low-frequency tracking (Molinaro et al., 2016; Di Liberto et al., 437 2018) while others demonstrate enhanced high-frequency tracking (Lehongre et al., 2011). Given 438 that slow and fast amplitude modulations are hierarchically nested in speech (Gross et al., 439 2013), atypical low-frequency temporal sampling could cascade to altered processing at faster 440 rates (Keshavarzi et al., 2025). According to this, impaired extraction of syllabic-level information 441 may require compensatory reliance on fine -grained temporal details (captured in alpha 442 frequencies in our stimulus material), resulting in less efficient reading. 443 In the current study, we cannot make causal inferences about whether enhanced cortical 444 tracking facilitates reading or whether better reading strengthens cortical tracking mechanisms. 445 Additional research is needed to determine whether training -induced changes in cortical 446 tracking would correspond to improvements in reading performance, or whether improvements 447 in reading would correspond to changed cortical tracking, which would provide evidence for 448 causal mechanisms. Additionally, reading proficiency comprises multiple component skills that 449 may exhibit distinct relationships with cortical tracking. Future investigations should examine 450 associations with specific subcomponents, including phonological decoding, reading fluency, 451 prosodic sensitivity, and reading comprehension. This approach would clarify which specific 452 aspects of reading ability are most closely linked to cortical tracking mechanisms. 453 Lateralised relationship between cortical tracking of speech and reading proficiency 454 The pattern of results found for the predictive role of speech tracking, whereby the 455 correlation of high -frequency cortical tracking with reading reaction times was left -lateralised, 456 and the correlation of low -frequency cortical tracking with reading was right -lateralised (Figure 457 6A), is reminiscent of the Asymmetric Sampling in Time (AST) theory (Poeppel, 2003). According 458 to the AST, the left hemisphere is proposed to be optimised for processing faster, fine -grained 459 temporal aspects of speech, whereas the right hemisphere is proposed to be optimised for 460 processing slower temporal features which unfold over time. While we did not find stronger 461 speech tracking of high - and low frequencies in left and right hemispheres, respectively 462 (consistent with previous studies, tracking was bilateral for both speech and music (Harding et 463 al., 2019), Figure 4), the left- and right-lateralised relationships with reading performance might 464 indicate that the processing of slower timescales in the right, and faster timescales in the left 465 hemisphere, is relevant for higher-level cognition such as reading. 466 Cortical tracking of speech is stronger than music at low frequencies 467 Previous work has shown that speech and music are tracked by cortical activity across 468 multiple timescales (Giraud & Poeppel, 2012; Doelling & Poeppel, 2015; Ding et al., 2017) . 469 Therefore, we expected cortical tracking across a wide range of channels and frequencies. Our 470

found significant cortical tracking for both speech and music across a wide spatial 471 distribution of channels and a broad spectral range. This extensive tracking pattern is consistent 472 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability with previous research (Te Rietmolen et al., 2024) and is thought to reflect the hierarchical 473 temporal structure of both domains, with different frequency bands capturing distinct structural 474 units, including prosody and rhythm. 475 Research has found mixed results on the strength of speech versus music tracking. Some 476 studies have shown that speech tracking is stronger than music tracking at low frequencies, 477 corresponding to syllables and phrases (der Nederlanden et al., 2020), while others find speech 478 tracking to be stronger at higher frequencies (Osorio & Assaneo, 2025). These discrepancies may 479 stem partly from methodological differences in how frequency ranges are analysed. Unlike 480 previous studies that averaged cortical tracking across frequency ranges (Harding et al., 2019), 481 we used a frequency-resolved mutual information approach, enabling a more fine -grained 482 comparison. This approach revealed that the relative strength of the cortical tracking is 483 frequency-dependent: speech tracking was stronger than music tracking for very low frequencies 484 (< 1 Hz), while music tracking was stronger than speech tracking across delta, theta and alpha 485 bands. The stronger speech tracking at very low frequencies may reflect the tracking of prosodic 486 phrases, which occur at rates below 1 Hz in natural speech . However, these frequency-specific 487 differences could also be influenced by the particular acoustic characteristics of our stimul i, 488 including differences in temporal regularity, rhythmic structure, and spectral content between 489 the selected speech and music. 490 Musical sophistication and reading enjoyment do not predict reading proficiency 491 Musical expertise has been associated with improved reading skills in children (Tierney & 492 Kraus, 2013; Garcia -de-Soria et al., 2025) , thought to be due to the positive effects of musical 493 expertise on brain plasticity and development (Olszewska et al., 2021) . However, evidence for 494 this relationship in adults is mixed. While some studies report positive associations between 495 musical training and phonological processing in adults (Pantaleo et al., 2024) , others find no 496 association between musical expertise and reading -related skills when controlling for general 497 cognitive abilities (Swaminathan et al., 2018) . Our results found that self -reported musical 498 sophistication did not significantly influence performance on the lexical decision task. This 499 suggests that potential transfer effects of musical training on language processing may be most 500 pronounced during development. The GMSI (Müllensiefen et al., 2013) provides a broad 501 assessment of musical sophistication , including self -reported engagement, training duration, 502 and perceptual abilities. However, it should be noted that more specialised measures targeting 503 specific perceptual skills such as pitch or rhythm discrimination may capture distinct 504 dimensions of musical ability not fully reflected in self -report indices, potentially explaining 505 differences in how musical experience relates to cortical tracking across studies. Furthermore, 506 observed differences between musicians and non -musicians may not reflect the effect s of 507 musical training itself, but rather stem from pre-existing aptitude differences that influence who 508 chooses to pursue musical training , leading to selection bias when comparing these groups 509 (Schellenberg & Lima, 2024). 510 Reading enjoyment has been shown to predict reading performance in children 511 (Retelsdorf et al., 2011), likely because greater enjoyment promotes increased reading practice, 512 which in turn strengthens reading skills (Jerrim et al., 2020). In contrast, the current study found 513 no significant relationship between reading enjoyment and lexical decision performance in 514 adults. This discrepancy may reflect developmental differences, as the influence of enjoyment 515 on reading ability may be more pronounced during childhood when literacy skills are still being 516 actively acquired. In adults who have already achieved functional literacy, neural processing 517 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability mechanisms indexed by cortical tracking may account for variance in reading performance 518 beyond that explained by behavioural factors such as reading enjoyment. 519 However, we did observe a significant positive relationship between age and reaction 520 times in the lexical decision task, with older participants in our sample responding more slowly. 521 The age range in the current study (19 – 28 years) was relatively narrow, however, this effect could 522 reflect subtle age-related slowing in reaction time even within young adulthood (Baudouin et al., 523 2004). 524

525 In conclusion, we show that cortical tracking of both speech and music predicts reading 526 performance in a frequency -dependent manner. Stronger delta -band tracking for both speech 527 and music was associated with faster lexical decision reaction times, whereas stronger alpha -528 band tracking for speech was associated with slower performance, suggesting that optimal 529 reading relies on differential engagement with temporal information across multiple timescales. 530 Importantly, this relationship was not modulated by stimulus type, musical sophistication or 531 reading enjoyment , suggesting that domain -general temporal processing mechanisms, rather 532 than stimulus-specific features, underlie the association between cortical tracking and reading 533 ability. However, causal mechanisms remain unclear, and future research should examine 534 whether training -induced changes in frequency -specific cortical tracking correspond to 535 improvements in reading outcomes. 536 537 Acknowledgments 538 We thank all participants who so generously gave their time and effort to take part in this 539 study. SCA is funded by the Scottish Graduate School of Social Science (SGSSS) for her doctoral 540 studies. AK is supported by the Medical Research Council [grant number MR/W02912X/1]. AK is 541 a member of the Scottish-EU Critical Oscillations Network (SCONe), funded by the Royal Society 542 of Edinburgh (RSE Saltire Facilitation Network Award to AK, Reference Number 1963). The 543 funders had no involvement in the study protocol, participant recruitment, data analysis, or 544 manuscript preparation. 545 Author contributions 546 S.C.A.: Conceptualisation, methodology, formal analysis, investigation, data curation, 547 writing—original draft, writing —review and editing, visualisation, funding acquisition. S.K.: 548 Investigation, writing – review. S.M.V.: Investigation, writing – review. A.K.: Conceptualisation, 549 methodology, formal analysis, investigation, software, writing – review and editing, supervision, 550 funding acquisition. 551 552 Conflict of interest 553 The authors declare no competing financial interests. 554 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability

555 Abrams, D. A., Nicol, T., Zecker, S., & Kraus, N. (2008). Right-hemisphere auditory cortex is 556 dominant for coding syllable patterns in speech. Journal of Neuroscience, 28(15), 3958-557 3965. 558 Ahissar, E., Nagarajan, S., Ahissar, M., Protopapas, A., Mahncke, H., & Merzenich, M. M. (2001). 559 Speech comprehension is correlated with temporal response patterns recorded from 560 auditory cortex. Proceedings of the National Academy of Sciences, 98(23), 13367-561 13372. 562 Anwyl-Irvine, A. L., Massonnié, J., Flitton, A., Kirkham, N., & Evershed, J. K. (2020). Gorilla in our 563 midst: An online behavioral experiment builder. Behavior research methods, 52(1), 388-564 407. 565 Araújo, J., Simons, B. D., Peter, V., Mandke, K., Kalashnikova, M., Macfarlane, A., Gabrielczyk, 566 F., Wilson, A., Di Liberto, G. M., & Burnham, D. (2024). Atypical low-frequency cortical 567 encoding of speech identifies children with developmental dyslexia. Frontiers in Human 568 Neuroscience, 18, 1403677. 569 Asano, R. (2022). The evolution of hierarchical structure building capacity for language and 570 music: a bottom-up perspective. Primates, 63(5), 417-428. 571 Atanasova, T., Gross, J., Rimmele, J. M., & Keitel, A. (2026). The involvement of endogenous 572 brain rhythms in speech processing. Neuroscience & Biobehavioral Reviews, 106568. 573 Baudouin, A., Vanneste, S., & Isingrini, M. (2004). Age-related cognitive slowing: The role of 574 spontaneous tempo and processing speed. Experimental aging research, 30(3), 225-575 239. 576 Bhide, A., Power, A., & Goswami, U. (2013). A rhythmic musical intervention for poor readers: A 577 comparison of efficacy with a letter‐based intervention. Mind, Brain, and Education, 578 7(2), 113-123. 579 Boersma, P. (2001). Praat, a system for doing phonetics by computer. Glot. Int., 5(9), 341-345. 580 Brainard, D. H., & Vision, S. (1997). The psychophysics toolbox. Spatial vision, 10(4), 433-436. 581 Cason, N., Astésano, C., & Schön, D. (2015). Bridging music and speech rhythm: Rhythmic 582 priming and audio–motor training affect speech perception. Acta psychologica, 155, 43-583 50. 584 Chalas, N., Daube, C., Kluger, D. S., Abbasi, O., Nitsch, R., & Gross, J. (2022). Multivariate 585 analysis of speech envelope tracking reveals coupling beyond auditory cortex. 586 NeuroImage, 258, 119395. 587 De Jong, N. H., & Wempe, T. (2009). Praat script to detect syllable nuclei and measure speech 588 rate automatically. Behavior research methods, 41(2), 385-390. 589 der Nederlanden, C. M. V. B., Joanisse, M. F., & Grahn, J. A. (2020). Music as a scaffold for 590 listening to speech: Better neural phase-locking to song than speech. NeuroImage, 214, 591 116767. 592 Di Liberto, G. M., Peter, V., Kalashnikova, M., Goswami, U., Burnham, D., & Lalor, E. C. (2018). 593 Atypical cortical entrainment to speech in the right hemisphere underpins phonemic 594 deficits in dyslexia. NeuroImage, 175, 70-79. 595 Ding, N., Patel, A. D., Chen, L., Butler, H., Luo, C., & Poeppel, D. (2017). Temporal modulations 596 in speech and music. Neuroscience & Biobehavioral Reviews, 81, 181-187. 597 Doelling, K. B., & Poeppel, D. (2015). Cortical entrainment to music and its modulation by 598 expertise. Proceedings of the National Academy of Sciences, 112(45), E6233-E6242. 599 Drakoulaki, K., Anagnostopoulou, C., Guasti, M. T., Tillmann, B., & Varlokosta, S. (2024). 600 Situating language and music research in a domain‐specific versus domain‐general 601 framework: A review of theoretical and empirical data. Language and Linguistics 602 Compass, 18(2), e12509. 603 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Flaugnacco, E., Lopez, L., Terribili, C., Zoia, S., Buda, S., Tilli, S., Monasta, L., Montico, M., Sila, 604 A., & Ronfani, L. (2014). Rhythm perception and production predict reading abilities in 605 developmental dyslexia. Frontiers in Human Neuroscience, 8, 392. 606 Garcia-de-Soria, M. C., Mathias, B., Keitel, A., & Klimovich-Gray, A. (2025). Benefits of music 607 training for learning to read: Evidence from cortical tracking of speech in children. 608 bioRxiv, 2025.2009. 2005.674218. 609 Giraud, A.-L., & Poeppel, D. (2012). Cortical oscillations and speech processing: emerging 610 computational principles and operations. Nature neuroscience, 15(4), 511-517. 611 Gorsuch, R. L., & Lehmann, C. S. (2010). Correlation coefficients: Mean bias and confidence 612 interval distortions. Journal of Methods and Measurement in the Social Sciences, 1(2), 613 52-65. 614 Goswami, U. (2007). Phonological representations for reading acquisition across languages. In 615 Single-Word Reading (pp. 78-97). Psychology Press. 616 Goswami, U. (2011). A temporal sampling framework for developmental dyslexia. Trends in 617 cognitive sciences, 15(1), 3-10. 618 Goswami, U., Huss, M., Mead, N., Fosker, T., & Verney, J. P. (2013). Perception of patterns of 619 musical beat distribution in phonological developmental dyslexia: Significant 620 longitudinal relations with word reading and reading comprehension. Cortex, 49(5), 621 1363-1376. 622 Gross, J., Hoogenboom, N., Thut, G., Schyns, P., Panzeri, S., Belin, P., & Garrod, S. (2013). 623 Speech rhythms and multiplexed oscillatory sensory coding in the human brain. PLoS 624 biology, 11(12), e1001752. 625 Haegens, S., & Golumbic, E. Z. (2018). Rhythmic facilitation of sensory processing: A critical 626 review. Neuroscience & Biobehavioral Reviews, 86, 150-165. 627 Haegens, S., Händel, B. F., & Jensen, O. (2011). Top-down controlled alpha band activity in 628 somatosensory areas determines behavioral performance in a discrimination task. 629 Journal of neuroscience, 31(14), 5197-5204. 630 Harding, E. E., Sammler, D., Henry, M. J., Large, E. W., & Kotz, S. A. (2019). Cortical tracking of 631 rhythm in music and speech. NeuroImage, 185, 96-101. 632 Ince, R. A., Giordano, B. L., Kayser, C., Rousselet, G. A., Gross, J., & Schyns, P. G. (2017). A 633 statistical framework for neuroimaging data analysis based on mutual information 634 estimated via a gaussian copula. Human brain mapping, 38(3), 1541-1573. 635 Jerrim, J., Lopez-Agudo, L. A., & Marcenaro-Gutierrez, O. D. (2020). Does it matter what children 636 read? New evidence using longitudinal census data from Spain. Oxford Review of 637 Education, 46(5), 515-533. 638 Keitel, A., Gross, J., & Kayser, C. (2018). Perceptually relevant speech tracking in auditory and 639 motor cortex reflects distinct linguistic features. PLoS biology, 16(3), e2004473. 640 Keitel, A., Pelofi, C., Guan, X., Watson, E., Wight, L., Allen, S., Mencke, I., Keitel, C., & Rimmele, 641 J. (2025). Cortical and behavioral tracking of rhythm in music: Effects of pitch 642 predictability, enjoyment, and expertise. Annals of the New York Academy of Sciences, 643 1546(1), 120-135. 644 Keshavarzi, M., Moore, B. C., & Goswami, U. (2025). Atypical low-frequency and high-frequency 645 neural entrainment to rhythmic audiovisual speech in adults with dyslexia. bioRxiv, 646 2025.2009. 2014.675562. 647 Koike, K. J., Hurst, M. K., & Wetmore, S. J. (1994). Correlation between the American Academy 648 of Otolaryngology—Head and Neck Surgery Five-Minute Hearing Test and Standard 649 Audiologic Data. Otolaryngology—Head and Neck Surgery, 111(5), 625-632. 650 Kotz, S. A., Ravignani, A., & Fitch, W. T. (2018). The evolution of rhythm processing. Trends in 651 cognitive sciences, 22(10), 896-910. 652 Kumagai, Y., Arvaneh, M., & Tanaka, T. (2017). Familiarity affects entrainment of EEG in music 653 listening. Frontiers in Human Neuroscience, 11, 384. 654 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Lehongre, K., Morillon, B., Giraud, A.-L., & Ramus, F. (2013). Impaired auditory sampling in 655 dyslexia: further evidence from combined fMRI and EEG. Frontiers in Human 656 Neuroscience, 7, 454. 657 Lehongre, K., Ramus, F., Villiermet, N., Schwartz, D., & Giraud, A.-L. (2011). Altered low-gamma 658 sampling in auditory cortex accounts for the three main facets of dyslexia. Neuron, 659 72(6), 1080-1090. 660 Lerdahl, F., & Jackendoff, R. (1983). An overview of hierarchical structure in music. Music 661 Perception, 229-252. 662 Lubineau, M., Watkins, C. P., Glasel, H., & Dehaene, S. (2024). Examining the Impact of Reading 663 Fluency on Lexical Decision Results in French 6th Graders. Open Mind, 8, 535-557. 664 Miyagawa, S., Berwick, R. C., & Okanoya, K. (2013). The emergence of hierarchical structure in 665 human language. Frontiers in Psychology, 4, 71. 666 Molinaro, N., Lizarazu, M., Lallier, M., Bourguignon, M., & Carreiras, M. (2016). Out‐of‐667 synchrony speech entrainment in developmental dyslexia. Human brain mapping, 668 37(8), 2767-2783. 669 Müllensiefen, D., Gingras, B., Stewart, L., & Musil, J. J. (2013). Goldsmiths Musical 670 Sophistication Index (Gold-MSI) v1. 0: Technical Report and Documentation Revision 671 0.3. In. 672 Nora, A., Rinkinen, O., Renvall, H., Arkkila, E., Smolander, S., Laasonen, M., & Salmelin, R. 673 (2024). Impaired cortical tracking of speech in children with developmental language 674 disorder. Journal of Neuroscience, 44(22). 675 Obleser, J., Eisner, F., & Kotz, S. A. (2008). Bilateral speech comprehension reflects differential 676 sensitivity to spectral and temporal features. Journal of Neuroscience, 28(32), 8116-677 8123. 678 Obleser, J., & Kayser, C. (2019). Neural entrainment and attentional selection in the listening 679 brain. Trends in cognitive sciences, 23(11), 913-926. 680 Oldfield, R. C. (1971). The assessment and analysis of handedness: the Edinburgh inventory. 681 Neuropsychologia, 9(1), 97-113. 682 Olszewska, A. M., Gaca, M., Herman, A. M., Jednoróg, K., & Marchewka, A. (2021). How musical 683 training shapes the adult brain: Predispositions and neuroplasticity. Frontiers in 684 neuroscience, 15, 630829. 685 Oostenveld, R., Fries, P., Maris, E., & Schoffelen, J.-M. (2011). FieldTrip: open source software 686 for advanced analysis of MEG, EEG, and invasive electrophysiological data. 687 Computational intelligence and neuroscience, 2011. 688 Osorio, S., & Assaneo, M. F. (2025). Anatomically distinct cortical tracking of music and speech 689 by slow (1–8Hz) and fast (70–120Hz) oscillatory activity. PLoS One, 20(5), e0320519. 690 Pantaleo, M. M., Arcuri, G., Manfredi, M., & Proverbio, A. M. (2024). Music literacy improves 691 reading skills via bilateral orthographic development. Scientific Reports, 14(1), 3506. 692 Patel, A. D. (2010). Music, language, and the brain. Oxford university press. 693 Patel, A. D. (2011). Why would musical training benefit the neural encoding of speech? The 694 OPERA hypothesis. Frontiers in psychology, 2, 142. 695 Peelle, J. E., & Davis, M. H. (2012). Neural oscillations carry speech rhythm through to 696 comprehension. Frontiers in Psychology, 3, 320. 697 Pelli, D. G., & Vision, S. (1997). The VideoToolbox software for visual psychophysics: 698 Transforming numbers into movies. Spatial vision, 10, 437-442. 699 Poeppel, D. (2003). The analysis of speech in different temporal integration windows: cerebral 700 lateralization as ‘asymmetric sampling in time’. Speech communication, 41(1), 245-701 255. 702 Power, A. J., Colling, L. J., Mead, N., Barnes, L., & Goswami, U. (2016). Neural encoding of the 703 speech envelope by children with developmental dyslexia. Brain and Language, 160, 1-704 10. 705 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Retelsdorf, J., Köller, O., & Möller, J. (2011). On the effects of motivation on reading 706 performance growth in secondary school. Learning and instruction, 21(4), 550-559. 707 Schellenberg, E. G., & Lima, C. F. (2024). Music Training and Nonmusical Abilities. Annu Rev 708 Psychol, 75, 87-128. https://doi.org/10.1146/annurev-psych-032323-051354 709 Scott, G. G., Keitel, A., Becirspahic, M., Yao, B., & Sereno, S. C. (2019). The Glasgow Norms: 710 Ratings of 5,500 words on nine scales. Behavior research methods, 51(3), 1258-1270. 711 Smith, Z. M., Delgutte, B., & Oxenham, A. J. (2002). Chimaeric sounds reveal dichotomies in 712 auditory perception. Nature, 416(6876), 87-90. 713 Sohoglu, E., Peelle, J. E., Carlyon, R. P., & Davis, M. H. (2012). Predictive top-down integration of 714 prior knowledge during speech perception. Journal of Neuroscience, 32(25), 8443-8453. 715 Swaminathan, S., Schellenberg, E. G., & Venkatesan, K. (2018). Explaining the association 716 between music training and reading in adults. Journal of Experimental Psychology: 717 Learning, Memory, and Cognition, 44(6), 992. 718 Symons, A. E., Dick, F., & Tierney, A. T. (2021). Dimension-selective attention and dimensional 719 salience modulate cortical tracking of acoustic dimensions. NeuroImage, 244, 118544. 720 Te Rietmolen, N., Mercier, M. R., Trébuchon, A., Morillon, B., & Schön, D. (2024). Speech and 721 music recruit frequency-specific distributed and overlapping cortical networks. Elife, 722 13, RP94509. 723 The Mathworks, I. (2025). MATLAB version: 25.1.0.2973910 (R2025a) Update 1. In The 724 MathWorks, Inc. https://www.mathworks.com 725 Thut, G., Nietzel, A., Brandt, S. A., & Pascual-Leone, A. (2006). α-Band electroencephalographic 726 activity over occipital cortex indexes visuospatial attention bias and predicts visual 727 target detection. Journal of neuroscience, 26(37), 9494-9502. 728 Tierney, A., & Kraus, N. (2013). Music training for the development of reading skills. Progress in 729 brain research, 207, 209-241. 730 Weems, S. A., & Zaidel, E. (2004). The relationship between reading ability and lateralized 731 lexical decision. Brain and Cognition, 55(3), 507-515. 732 Yao, B., Keitel, A., Bruce, G., Scott, G. G., O'Donnell, P. J., & Sereno, S. C. (2018). Differential 733 emotional processing in concrete and abstract words. Journal of Experimental 734 Psychology: Learning, Memory, and Cognition, 44(7), 1064. 735 736 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability Supplemental material 737 Correlation between cortical tracking and word trials in the lexical decision task 738 To test whether envelope tracking during listening to the story and music predicted participants’ 739 behavioural performance in the lexical decision task, we correlated the MI values per electrode 740 and frequency with participants’ median reaction time across word trials. 741 For the story condition, we found one negative cluster, indicating that participants who showed 742 stronger envelope tracking to speech at low frequencies had faster reaction times in the lexical 743 decision task (Figure S1A(i), highlighted in red). This negative cluster (Figure S1A(i)) at low 744 frequencies ( 0.83 Hz - 1.66 Hz), significantly predicted reaction times to words in the lexical 745 decision task (Cohen’s dpeak = 1.79, pcluster < .001, 10 electrodes). We also found a positive cluster 746 at higher frequencies (highlighted in black in Figure S1A(i) spanning 10.05 Hz – 13.26 Hz, (Cohen’s 747 dpeak = 3.11, pcluster < .001, 3 electrodes), indicating that participants who showed stronger 748 envelope tracking to speech had slower reaction times to words in the lexical decision task. 749 For the music condition, we found one negative cluster, similarly indicating that participants who 750 showed stronger envelope tracking to music at low frequencies had faster reaction times in the 751 lexical decision task (Figure S1A(ii). The negative cluster spanned 0.89 Hz - 1.44 Hz, also 752 predicting reaction times (Cohen’s dpeak = 2.32, pcluster < .001, 12 electrodes). We also found one 753 positive cluster spanning 3.55 – 4.37 Hz, indicating that participants who showed stronger 754 envelope tracking to music had slower reaction times in the lexical decision task (Cohen’s dpeak 755 = 1.24, pcluster < .001, 3 electrodes). 756 Correlation between cortical tracking and nonword trials in the lexical decision task 757 To test whether envelope tracking during listening to the story and music predicted participants’ 758 behavioural performance in the lexical decision task, we correlated the MI values per electrode 759 and frequency with participants’ median reaction time across nonword trials. 760 For the story condition we found three clusters, two negative and one positive. The first negative 761 cluster spanned 1.02 Hz - 1.66 Hz (Figure S1B(i) outlined in black), that predicted reaction times 762 in the lexical decision task (Cohen’s dpeak = 1.07, pcluster < .001, 3 electrodes). The second negative 763 cluster at 7.61 Hz – 8.75 Hz also predicted reaction times in the lexical decision task (Cohen’s 764 dpeak = 1.30, pcluster < .001, 3 electrodes, Figure S1B(i) outlined in yellow), indicating that 765 participants who showed stronger envelope tracking to speech at low frequencies had faster 766 reaction times in the lexical decision task. We also found one positive cluster (Figure S1B(i) 767 outlined in green), indicating that participants who showed stronger envelope tracking to speech 768 at low frequencies had slower reaction times in the lexical decision task. The positive cluster 769 spanned 8.75 Hz – 13.26 Hz (Cohen’s dpeak = 1.93, pcluster < .001, 6 electrodes). 770 For the music condition, we found one negative cluster in frontal electrodes at low frequencies 771 (0.89 Hz - 1.35 Hz, Figure S1B(ii)) that predicted reaction times in the lexical decision task 772 (Cohen’s dpeak = 1.84, pcluster < .001, 11 electrodes). This indicates that participants who showed 773 stronger envelope tracking to music at low frequencies had faster reaction times in the lexical 774 decision task. 775 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint Allen et al. Tracking of speech and music predicts reading ability 776 Figure S1. Relationship between cortical tracking and performance on the lexical decision task, quantified 777 through reaction times for word and nonword trials separately. (A) Channel -by-frequency heatmap of the 778 correlation between cortical tracking of speech and participants’ reaction times to word trials. Two 779 clusters were found for speech, one negative cluster at low frequencies, and one positive cluster at higher 780 frequencies. Two clusters were found for music: one negative cluster indicating that stronger tracking of 781 music correlated with faster reaction times to words in the lexical decision task, and one positive cluster, 782 indicating that stronger tracking of music at higher frequencies correlated with slower reaction times to 783 words. (B) Channel -by-frequency heatmap of the correlation between cortical tracking of music and 784 participants' reaction times to nonword trials. Three clusters were found for speech, two negative clusters 785 indicating that stronger tracking of speech correlated with faster reaction times to nonwords in the lexical 786 decision task, and one positive cluster at higher frequencies (outlined in green), indicating that stronger 787 tracking of speech correlated with slower reaction times to nonwords. One negative cluster was found for 788 music, indicating that stronger tracking of music correlated with faster reaction times to nonwords in the 789 lexical decision task. 790 791 .CC-BY-NC 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted February 18, 2026. ; https://doi.org/10.64898/2026.02.18.706526doi: bioRxiv preprint