Optimizing parameters for multiband ABRs to continuous speech

README ------ Details related to access to the data ------------------------------------- Please contact the following author for further information: Melissa Polonenko(email: [e-mail]) Overview -------- This is the dataset for the paper: "Optimal parameters for measuring multiband auditory brainstem responses to continuous speech" Polonenko MJ & Eisenreich BR (2026), with citation listed below. Trends in Hearing: Melissa J. Polonenko, Benjamin R. Eisenreich, Trends in Hearing, 2026; https://doi.org/10.1177/23312165261465722 BioRxiv: Melissa J. Polonenko, Benjamin R. Eisenreich, bioRxiv 2025.12.24.696406; doi: https://doi.org/10.64898/2025.12.24.696406 Exp1_phases ----------- This is the "phases" dataset for the paper Multiband auditory brainstem responses (ABRs) were derived to continuous peaky speech from one narrator with his original fundamental frequency (125 Hz). Data was collected from October to December 2022. Aim: Determine if peaky speech made with chirp-phase profile evokes larger ABRs than zero-phase profile. Based on: 1) previous work of click vs chirp stimuli for ABRs. 2) computational modeling with 11 stories with a range of natural f0s to zero- and CE-chirp phase profiles. The details of the experiment can be found in the bioRxiv pre-print and article cited above. Stimuli: multiband peaky speech (using audiological bands, dichotic) for 1 talker (f0 = 118 Hz) at the natural f0, with zero- or chirp-phase profile. 360 x 10 s trials each of the 2 phase profiles for a total of 720 trials (120 minutes or 2 hours, 60 min each) The code for analyses is available in the code/exp1_phases subfolder here, and on Github: https://github.com/polonenkolab/multibandABR_optimal_params Format ------ The dataset is formatted according to the EEG Brain Imaging Data Structure. It includes EEG recording from participant 01 to 15 in raw brainvision format (3 files: .eeg, .vhdr, .vmrk) and stimuli files in format of .hdf5. The stimuli files contain the audio ('x_play_band_zero' and 'x_play_band_chirp'), and regressors for the deconvolution ('pulse_inds' are the pulse indices for the multiband ABRs, 'fake_pulse_inds' are the pulse indices for creating the "common component" that is subtracted to reveal the frequency-specific ABRs). Generally, you can find detailed event data in the .tsv files and descriptions in the accompanying .json files. Raw eeg files are provided in the Brain Products format. Participants ------------ 17 participants, mean ± SD age of 20.4 ± 1.3 years (19-23 years) NOTE: do NOT use subjects 1 (no responses) and 13 (last 1 hr of data not saved) Thus, use 15 participant data. Inclusion criteria: 1) Age between 18-40 years 2) Normal hearing: audiometric thresholds <25 dB HL from 500 to 8000 Hz 3) Speak English as their primary language Please see participants.tsv for more information. Apparatus --------- Participants sat in a darkened sound-isolating booth and rested while listening to the audiobooks, although they were not required to pay attention to the stories. Stimuli were presented at an average level of 65 dB SPL and a sampling rate of 48 kHz through ER-2 insert earphones plugged into an RME Digiface USB digital sound card. Custom python scripts using expyfun were used to control the experiment and stimulus presentation. Details about the experiment ---------------------------- For a detailed description of the task, see Polonenko & Eisenreich (2026) and the supplied `task-phases_eeg.json` file. The 2 phase conditions were randomly interleaved. Trigger onset times are not corrected for the delay of the insert earphones, but the tsv files have both the uncorrected and corrected samples. Triggers with values of "1" were recorded to the onset of the 10 s audio, and shortly after triggers with values of "4" or "8" were stamped to indicate the which of the 2 conditions was played, the overall trial number, and the stimulus file number. This was done by converting the decimal trial number to bits, denoted b, then calculating 2 ** (b + 2). We've specified these trial numbers and more metadata of the events in each of the '*_eeg_events.tsv" file, which is sufficient to know which trial corresponded to which type of stimulus and which file - e.g., stimuli/exp1_phases/wizardofoz/wizardofoz_original_0001.hdf5. Exp2_f0s -------- This is the "f0s" dataset for the paper Multiband auditory brainstem responses (ABRs) were derived to continuous peaky speech from two narrators (stories) with different fundamental frequencies (f0s) that were also shifted down to an average f0 of 100 Hz. Data was collected from January to April 2023. Aim: Evaluate the f0 effect for measured multiband ABRs, using 2 talkers with different natural f0s (one <170 Hz and one >170 Hz) and both shifted down to an average f0 of 100 Hz. Based on: 1) previous work (Polonenko & Maddox, 2021, 2024) that showed for broadband ABRs, narrators with lower f0s produce larger ABRs, with an effect similar to the stimulation rate effect in click ABRs. 2) computational modeling with systematic shifting of 11 stories with a range of natural f0s to each mean f0 from 80 to 130 Hz in 10 dB steps. The details of the experiment can be found in the bioRxiv pre-print and article cited above. Stimuli: multiband peaky speech (using audiological bands, dichotic) for two talkers (f0 = 118 Hz and 180 Hz) at their natural f0s and shifted to 100 Hz, 210 x 10 s trials each of the 4 narrator-f0 combo for a total of 840 trials (2 hours 20 minutes, 35 min each) NOTE: f0s used: original f0s (118 and 180 Hz) and f0s shifted to 100 Hz The code for analyses is available in code/exp2_f0s subfolder herein and on Github: https://github.com/polonenkolab/multibandABR_optimal_params Format ------ The dataset is formatted according to the EEG Brain Imaging Data Structure. It includes EEG recording from participant 01 to 30 in raw brainvision format (3 files: .eeg, .vhdr, .vmrk) and stimuli files in format of .hdf5. The stimuli files contain the audio ('x_play_band_chirp_CE'), and regressors for the deconvolution ('pulse_inds' are the pulse indices for the multiband ABRs, 'fake_pulse_inds' are the pulse indices for creating the "common component" that is subtracted to reveal the frequency- specific ABRs). Generally, you can find detailed event data in the .tsv files and descriptions in the accompanying .json files. Raw eeg files are provided in the Brain Products format. Participants ------------ 30 participants, mean ± SD age of 20.3 ± 2.0 years (18-29 years) NOTE: removed participants sub-exp223 and sub-exp230 from analysis due to poor responses (there were problems noted during the recording). Participants sub-exp206, sub-exp208, sub-exp217 were also removed from the paper's analysis due to noisy responses. Thus n = 25 total used in the final analysis. Inclusion criteria: 1) Age between 18-40 years 2) Normal hearing: audiometric thresholds 20 dB HL or better from 500 to 8000 Hz 3) Speak English as their primary language Please see participants.tsv for more information. Apparatus --------- Participants sat in a darkened sound-isolating booth and rested while listening to the audiobooks, although they were not required to pay attention to the stories. Stimuli were presented at an average level of 65 dB SPL and a sampling rate of 48 kHz through ER-2 insert earphones plugged into an RME Digiface USB digital sound card. Custom python scripts using expyfun were used to control the experiment and stimulus presentation. Details about the experiment ---------------------------- For a detailed description of the task, see Polonenko & Eisenreich (2026) and the supplied `task-f0s_eeg.json` file. The 4 conditions (2 narrators x 2 f0s) were randomly interleaved for each block of trials (i.e., for trial 1, the 4 conditions were randomized). Trigger onset times are not corrected for the delay of the insert earphones, but the tsv files have both the uncorrected and corrected samples. Triggers with values of "1" were recorded to the onset of the 10 s audio, and shortly after triggers with values of "4" or "8" were stamped to indicate the which of the 4 conditions was played, the overall trial number, and the stimulus file number. This was done by converting the decimal trial number to bits, denoted b, then calculating 2 ** (b + 2). We've specified these trial numbers and more metadata of the events in each of the '*_eeg_events.tsv" file, which is sufficient to know which trial corresponded to which type of stimulus (toto or wizardofoz story), which f0 (100 Hz or original f0), and which file - e.g., stimuli/exp2_f0s/toto_shifted/toto_shifted_0001.hdf5 for the toto story with 100 Hz f0. Exp3_f0s2 --------- This is the "f0s2" dataset for the paper Multiband auditory brainstem responses (ABRs) were derived to continuous peaky speech from two narrators (stories) with fundamental frequencies (f0s) that were shifted down to an average f0 of 90 Hz. Data was collected in June 2023. Aim: Determine if the f0 shifted down to 90 Hz gives good ABRs across stories. Based on: 1) previous work (Polonenko & Maddox, 2021, 2024) that showed for broadband ABRs, narrators with lower f0s produce larger ABRs, with an effect similar to the stimulation rate effect in click ABRs. 2) computational modeling with systematic shifting of 11 stories with a range of natural f0s to each mean f0 from 80 to 130 Hz in 10 dB steps. 3) f0s study that showed stories shifted to 100 Hz gave larger ABRs. The details of the experiment can be found in the bioRxiv pre-print and article cited above. Stimuli: multiband peaky speech (using audiological bands, dichotic) for 3 talkers (f0s = 118, 132, 155 Hz) at their natural f0s and shifted to 90 Hz, 210 x 10 s trials each of the 3 narrators for a total of 630 trials (105 minutes or 1.75 hours, 35 min each) The code for analyses is available in the code/exp3_f0s2 subfolder herein and on Github: https://github.com/polonenkolab/multibandABR_optimal_params Format ------ The dataset is formatted according to the EEG Brain Imaging Data Structure. It includes EEG recording from participant 01 to 15 in raw brainvision format (3 files: .eeg, .vhdr, .vmrk) and stimuli files in format of .hdf5. The stimuli files contain the audio ('x_play_band_chirp_CE'), and regressors for the deconvolution ('pulse_inds' are the pulse indices for the multiband ABRs, 'fake_pulse_inds' are the pulse indices for creating the "common component" that is subtracted to reveal the frequency- specific ABRs). Generally, you can find detailed event data in the .tsv files and descriptions in the accompanying .json files. Raw eeg files are provided in the Brain Products format. Participants ------------ 15 participants, mean ± SD age of 27.7 ± 10.2 years (19-57 years) NOTE: all participants included in final analysis. Inclusion criteria: 1) Age between 18-60 years 2) Normal hearing: audiometric thresholds 20 dB HL or better from 500 to 8000 Hz 3) Speak English as their primary language Please see participants.tsv for more information. Apparatus --------- Participants sat in a darkened sound-isolating booth and rested while listening to the audiobooks, although they were not required to pay attention to the stories. Stimuli were presented at an average level of 65 dB SPL and a sampling rate of 48 kHz through ER-2 insert earphones plugged into an RME Digiface USB digital sound card. Custom python scripts using expyfun were used to control the experiment and stimulus presentation. Details about the experiment ---------------------------- For a detailed description of the task, see Polonenko & Eisenreich (2026) and the supplied `task-f0s2_eeg.json` file. The 3 conditions (3 narrators) were randomly interleaved for each block of trials (i.e., for trial 1, 3 stories randomized). Trigger onset times are not corrected for the delay of the insert earphones, but the tsv files have both the uncorrected and corrected samples. Triggers with values of "1" were recorded to the onset of the 10 s audio, and shortly after triggers with values of "4" or "8" were stamped to indicate the which of the 3 conditions was played, the overall trial number, and the stimulus file number. This was done by converting the decimal trial number to bits, denoted b, then calculating 2 ** (b + 2). We've specified these trial numbers and more metadata of the events in each of the '*_eeg_events.tsv" file, which is sufficient to know which trial corresponded to which type of stimulus and which file - e.g., stimuli/exp3_f0s2/wizard/wizard_shifted_0001.hdf5.