Data on electrophysiological responses and psychological reactions to 3D audio scenes and unpleasant International Affective Picture System images stimulation

Abstract

The strong potential of three-dimensional (3D) audio to elicit sensory and cognitive responses suggests its possible integration into rehabilitation and palliative treatments, particularly for anxiety. However, the benefits of using 3D audio to address this emotional state are still emerging as a research area. This work presents data collected from an experiment designed to evaluate the effect of binaural 3D audio on emotional responses, and specifically anxiety. The dataset includes demographic information, psychological scores, electrophysiological signals, preprocessing codes, and the audiovisual media used during the study. Sixty healthy Mexican participants observed aversive images in random order to induce a transient state of anxiety. Subsequently, they listened to one of five audio scenes intended to modulate their emotional responses. Electrocardiographic and electrodermal activity were recorded throughout the experiment, and behavioral data were gathered using self-report questionnaires. This dataset could be reused to study the relationship between psychological and autonomic nervous system responses, and to explore the potential of spatial auditory stimulation for anxiety modulation and mental health applications.

Background & Summary

The integration of binaural 3D audio, a form of auditory extended reality, into psychological research has received growing attention over recent decades due to its ability to stimulate sensory and cognitive processes through spatial acoustic information^1–3. Prior studies have shown that emotional arousal elicits measurable autonomic nervous system responses, typically captured through physiological indicators such as heart rate variability and skin conductance^4–6, as well as through self-report instruments assessing valence, arousal, dominance, and anxiety^7,8. Notably, unpleasant and arousing scenarios have been associated with increased cardiac and electrodermal activity⁹. However, despite expanding interest in audio-based technologies, their potential therapeutic use for reducing emotional symptoms and physiological responses related to anxiety remains underexplored.

To address this gap in knowledge, we collected data for a study comparing the effects of four binaural 3D audio soundscapes on anxiety reduction. This dataset includes psychological and physiological responses to evaluate the impact of aversive images from the International Affective Picture System (IAPS)¹⁰, used to induce an anxious state, and 3D auditory scenes, comparing these with self-help techniques recommended by the World Health Organization (WHO)¹¹. Data were gathered from 60 healthy Mexican participants with low trait anxiety levels, and include responses to the Self-Assessment Manikin test (SAM)¹², the Spanish version of the State-Trait Anxiety Inventory (STAI)¹³, as well as Electrocardiographic (ECG) and Electrodermal Activity (EDA) recordings. In addition to participant data, the dataset includes the four spatial audio files used in the study, rendered in 7th-order ambisonics format, and a monophonic audio file from the WHO audio guide. The Ethics Committee in Clinical Research (Protocol ID: P000858-3DAUDIO-ANX-TEC24-CEIC-CR001, IRB committee number: COFEPRIS 20 CI 19 039 002) and the Research Committee (Protocol ID: P000858-3DAUDIO-ANX-TEC24-CI-CR001, IRB committee number: CONBIOETICA 19 CEI 011-2016-10-17) of Tecnologico de Monterrey granted approval for conducting the study and sharing the data collected.

The outcomes of this dataset may support the integration of spatial audio stimuli and tools into psychological and palliative interventions aimed at reducing anxiety and improving mental health. Other researchers could use this dataset to extract electrophysiological features of the autonomic nervous system and investigate their association with emotional responses elicited by aversive visual stimuli and auditory scenes across different emotional contexts. What differentiates this dataset from previous works is the inclusion of high-quality audio files, developed under the supervision of audio professionals and specifically designed based on the emotional perception of the Mexican population, in accordance with the recommendations of the International Telecommunication Union (ITU)¹⁴ and the standards of the International Affective Digitized Sounds (IADS)¹⁵. Additionally, the soundscapes were presented in a validated 3D audio environment with head-tracking.

Figure 1 presents the study design for data collection. Participants completed two visits. The first visit aimed at screening and included participants with neurological, psychological, and auditory health within the study criteria. During the second visit, the experiment was conducted by collecting state anxiety and SAM scores at three time points: resting state, after visual stimulation, and after auditory stimulation. Electrophysiological signals were recorded before the experiment, during visual stimulation, and during auditory stimulation. Visual stimuli were presented on a computer monitor, while audio stimuli were delivered through professional audio production headphones with a head tracker.

Fig. 1.

Study design: from inclusion assessments to stimuli exposure. This study included two sessions, one of them lasting 45 min and the other lasted approximately 35 min.

Methods

Data collection sample

This dataset corresponds to the information of 60 healthy Mexican participants. The sample was balanced for three factors: 1) Sex: 30 males, 30 females; 2) age range according to the WHO classification: 20 teenagers (18-19 years old), 20 young adults (20–24 years old), and 20 adults (25–60 years old); and 3) audio scenes: 12 individuals for each condition. The age range was 18–44 years old, with a mean age of 24 years. It consisted mainly of graduate and postgraduate students from Tecnologico de Monterrey (47 participants). All individuals reported to be non-expert listeners. Only participants who reported trait anxiety levels below decile 8, and healthy auditory and neurological health were considered for the study. Additionally, the dataset includes the auditory stimuli presented to participants, consisting of four 3D audio scenes: 1) urban park (PK), 2) electronic music concert (EL), 3) local market (MK), and 4) war zone (WZ). There is a fifth audio consisting of the adaptation of the audio exercises that were based on the WHO audio guide, “Doing What Matters in Times of Stress” (DWM)¹¹. This recording is presented in monophonic format and served as the stimulus for the control group. The participants were recruited through institutional and commercial social networks. They were given an informed consent form before the experiment, agreed to participate in the study, and provided authorization to share the collected data. Each participant was assigned an identification number (ID) to exclude personal information, such as name and contact details, from the dataset.

Demographic, psychological and audiological data collection

Dataset includes demographic information such as ID, sex, age, age class, university enrollment, assigned audio group, STAI level scores and deciles, SAM test scores, and Pure Tone Audiometry (PTA) hearing levels. The anxiety levels were measured on 4-point scales from zero to three, and the emotional dimensions were graded on 9-point scales from one to nine. Hearing levels were assessed by means of pure-tone audiometry (PTA) in the range 125 – 8k Hz, and neurological health was evaluated using the Neurological Assessment from the Institute of Neuroscience at the University of Guadalajara.

Data acquisition

ECG and EDA physiological signals were recorded simultaneously using the g.USBamp RESEARCH amplifier system (g.tec medical engineering GmbH, Austria) with a sampling frequency of 256 Hz. ECG signals were acquired with the g.GAMMAbox interface connected to the g.USBamp RESEARCH, using three g.GAMMAclip connectors: an active input, a reference, and a ground (g.GAMMAclipGND). EDA signals were recorded using Galvanic Skin Response (GSR) sensors connected to the g.GSRsensor driver and to the g.USBamp RESEARCH. Additionally, the OpenVibe software 3.3.1 (Inria Hybrid Team, France) including the Designer and Acquisition 64-bit servers, was used for real-time acquisition and visualization.

Visual stimuli

For state anxiety induction, 56 images from the IAPS database that had been previously validated for the Mexican population were used¹⁶. Only aversive images (low valence, score < 5; high arousal, score > 5) were included in the experiment. MATLAB R2020b (The Mathworks, Inc., USA) was used to randomize the order presentation of images, and to create a video of the images, each one presented for six seconds. The rationale for this randomization was to induce anxiety through anticipation^17–19. Participants were aware that the images were intended to elicit a transient anxiety period, but they did not know which specific type of image would be displayed next. The video was 5 min 36 s long.

Audio stimuli

3D audio scenes

The audio scenes were developed by project co-investigators with academic degrees and professional experience in audio production engineering. For the 3D audio scenes, high-quality files from Freesound repository (Universitat Pompeu Fabra, Spain) and in-house audio libraries were selected to ensure audio fidelity. The DT990pro headphones (Beyerdynamic GmbH & Co. KG, Germany) and a Headtracker 1 (Supperware Ltd., UK) device facilitated real-time monitoring with head tracking. For the EL category, a multitrack remake of “Rasputin” by Majestic was bought and downloaded from the internet for optimal quality. The unmixed, full multitrack version of the song EL used in this study is available for purchase at https://www.logictemplates.com/lgten/rasputin-by-majestic-boney-m.

The software REAPER v6.58 (Cockos Incorporated, USA), utilizing MultiEncoder, ALLRADecoder plugins from the IEM Plug-in suite (Institute of Electronic Music and Acoustics, Austria), the 6-band stereo renaissance equalizer plugin (Waves Inc, USA) and Virtuoso v1.2.1 binauralizer (Applied Psychoacoustics Lab, University of Huddersfield Enterprises Limited, UK) were employed during the sessions. To improve computational efficiency for experiments, loudspeaker channel bounces were prepared in post-mixing. The files incorporated in 3D audio scenes had a sampling rate of at least 44.1k Hz, and 16-bit depth.

The 3D audio narrative structure was influenced by post-production techniques akin to short film audio design. Final mixes were adapted to a 26p Lebedev virtual loudspeaker array. Figure 2 shows an image of the workflow for audio design.

Fig. 2.

Computer screenshot of the mixing session for designing 7th-order ambisonics audio scenes. All applications presented in the image were used in real time to monitor the mix.

WHO self-help techniques audio guide

The transcription and adaptation of the audio exercises were based on the World Health Organization’s guide, “Doing What Matters in Times of Stress.” This guide provides five self-help techniques: Grounding, Detaching, Acting According to your Values, Being Kind, and Making Space, supported by eight recordings lasting 38 minutes and 11 seconds. To match the 5 min 36 s exposure time, the script form of the audio guides was adapted. Adjustments included omitting exercises involving bodily movements to avoid interference with biosignal capture and ensuring acoustic clarity by addressing issues like room reverberation and speaker intelligibility.

The DWM audio guide was re-recorded in the male voice of a co-investigator, synchronized to the duration of aversive image presentations, and enhanced with professional-grade audio equipment. For the recording, the equipment, audio processing settings, and narration parameters are summarized in Table 1.

Table 1.

Recording equipment and audio parameters.

Component	Specification
Microphone	Sennheiser MD 421-II
Audio interface	Behringer FCA610
Digital Audio Workstation	REAPER v6.58
Recording parameters	48k Hz sample rate, 24-bit depth
Equalizer	ReaEQ equalizer with a high-pass filter at 100 Hz and a 2.7 dB gain boost at 218.8 Hz
Compressor	ReaComp compressor, −45 dB threshold, 2:1 ratio, optimized attack and release settings.
Narration	Male voice, 5 min 36 s total duration; aprox. 3 s silence between technique indications; pacing adjusted to fit content within duration

The table details the hardware, software, and audio processing settings, as well as narration parameters, employed in the recording of the audio guide.

Experimental procedure

The study consisted of two individual sessions. The first session included study explanation, consent signing, neurological and auditory assessments, and the trait anxiety inventory from STAI. Participants were randomly assigned to one of five treatments: WZ (low valence, high arousal), PK (high valence, low arousal), EL (high valence, high arousal), MK (low valence, low arousal), or DWM (control group), with allocation balanced by sex and age.

During the second session, participants completed the STAI state anxiety inventory and SAM. The researcher then cleaned specific areas of the participant’s skin with 70% alcohol for electrode placement: mid-arms, left clavicle, and inner surfaces of the left index and middle fingers. ECG electrodes followed Khan’s²⁰ modified Einthoven triangle, that is, the anode electrode on the right mid-arm below the V4 line, the cathode on the left mid-arm, and the ground on the left clavicle, while GSR electrodes were placed on the left index and middle fingers²¹. Figure 3 shows the electrode placement on participants and the electrophysiological acquisition systems used.

Fig. 3.

Electrode montage on participants and the acquisition systems used during the experiment. (A) Real-time visualization of ECG and EDA signals. (B) EDA sensors. (C) ECG electrodes. V1: active input; REF: reference electrode; GND: ground electrode. Created in BioRender. Santiago Molina, S. (2025) https://BioRender.com/21ln4zo.

Participants sat with their feet flat and palms resting on the table. Then, they wore headphones with a head tracker and were familiarized with 3D audio through an example that demonstrated head-tracking movements, allowing them to rotate their head in the pitch, yaw and roll dimensions of movement. They were instructed not to make fast and abrupt movements, but to perform natural rotations of the head to reduce artifacts in the recordings. After the explanation, they were recorded in a resting state with their eyes closed for three minutes. Subsequently, they watched a video with 56 aversive images from the IAPS, each randomly displayed for six seconds to induce state anxiety through anticipation, as it has been previously demonstrated^17–19. They then completed the SAM and state anxiety tests after exposure. Next, participants listened to a 3D audio scene or a guided self-help recording with their eyes closed for 5 min 36 s. Thereafter, they completed the SAM and state anxiety tests after the treatment.

Finally, electrodes and headphones with the head tracker were removed, and participants were asked about any discomfort caused by the images or audio. Those reporting discomfort were directed to the Emotional Counselling department of Tecnologico de Monterrey for individual support and following.

Data Records

The data presented in this work corresponds to 60 Mexican participants, and five audio scenes, four designed in ambisonics 3D audio format, and one designed in monophonic format. Additionally, information on the presented aversive images is provided. Figure 4 presents data organization in the repository. Personal information such as name, email and phone number of participants is not included in the data. This dataset was uploaded to the Research Data Hub of Tecnologico de Monterrey²² (link to repository: 10.57687/FK2/LCB08J).

Fig. 4.

Data organization in the Research Data Hub repository. Data available are neurophysiological data of ECG and EDA, audio files used for auditory stimulation, code script for video generation, preprocessing codes for biosignals, and data specifications of participants and sound scenes.

AudioFiles

This ZIP folder has five subfolders with WAV files with 5 min 36 s audio scenes. The files are given in the following playback modalities: (1) 26-channel 7^th order ambisonics mixes, (2) headphone calibrated virtual room binaural mixes, and (3) anechoic binaural mixes. Only a 15 s fragment of EL is included as a preview due to intellectual property of the actual song. The binaural audio files were normalized to −23 Loudness Units. For DWM, only one audio file is available since it was implemented in the control group, without 3D audio spatialization.

DataSpecs

This file has demographic data of participants in the following categories: (1) ID, (2) sex (3) age, (4) age class, (5) university enrolment, (6) audio group, (7) STAI trait anxiety score, (8) STAI trait anxiety decile, (9) STAI state anxiety score during resting state, (10) STAI state anxiety decile during resting state, (11) STAI state anxiety score after aversive image stimulation, (12) STAI state anxiety decile after aversive image stimulation, (13) STAI state anxiety score after auditory stimulation, (14) STAI state anxiety decile after auditory stimulation, and (15) hearing levels for the left and right ears in the range 125 – 8k Hz assessed by means of PTA. It also includes the descriptions and links to the open-access audio assets used in the creation of the 3D audio scenes.

Dictionary

This file explains several variables contained in the files described above in terms of concept, data type, description, units, value range, and the file in which each variable can be found.

IAPSimages

This file has the ID of the aversive images presented according to the numeration in the IAPS database. They were selected based on the affective responses of the Mexican population measured in a previous study. The content of these images includes body fluids, insects, animals, suicide tendencies, violence and internal organs.

PreprocessedData

This ZIP folder has the preprocessed version of electrophysiological recordings. It has individual subfolders for each biosignal. ECG and EDA signal are separated for each condition: 1) Resting state (Bl), 2) IAPS, and 3) audio scene (WZ, PK, EL, and MK). For example, the files “02-Bl-ECG-clean” indicate the preprocessed ECG recording of participant 02 during resting state condition. The SET and FDT files are needed to open the recordings in EEGLAB. It also includes the CSV file versions. Table 2 presents the acquisition and preprocessing parameters of the preprocessed files.

Table 2.

ECG and EDA acquisition and preprocessing parameters.

Parameter	ECG	EDA
Sampling rate	256 Hz
Electrode montage	Khan’s modified Einthoven triangle: the anode electrode on the right mid-arm below the V4 line, the cathode on the left mid-arm, and the ground on the left clavicle.	GSR electrodes were placed on the left index and middle fingers
Preprocessing steps in EEGLAB	(1) DC offset removal by signal mean subtraction; (2) artifact segment removal using Artifact Subspace Reconstruction method; and (3) FIR band-pass filtering (0.3–30 Hz).	(1) DC offset removal by mean subtraction; (2) Gaussian low-pass filtering at 0.1 Hz; (3) minimum value added to avoid negative units after DC offset removal; and (4) downsampling to 4 Hz for smoothing.
Artifact handling	Segments within artifact windows removed via ASR; some events excluded due to overlap with artifact windows (latencies retrievable from raw files).	Segments with abrupt changes removed; smoothing performed during downsampling.
Preprocessing software	EEGLAB v2023.0 on MATLAB R2020b

Each biosignal includes key details relevant for technical validation.

PreprocFunctions

This ZIP folder contains the MATLAB scripts for preprocessing the ECG and EDA signals, following the pipeline described in Table 2.

RawData

This ZIP folder has individual subfolders of participants with their assigned ID. Inside each folder, EDF files are shown corresponding to ECG and EDA recordings. Each EDF file has the recording of two channels. Channel 1 presents the ECG signal, and channel 2 presents the EDA signal. Stimulation code “OVTK_StimulationId_Number_07” in EDF files enabled the video of images while stimulation code “OVTK_StimulationId_Number_10” enabled the assigned audio scene. Stimulation code “condition 2” started the images or the audio scenes, depending on the stimuli condition. It also includes the CSV file versions.

README

These Markdown and PDF files describe the structure of the dataset and provide guidance on how to preprocess the electrophysiological recordings and play the audio files.

VideoGenerator

This file refers to the MATLAB code used to generate the video for randomized image presentation according to the IAPSimages file. The presentation order was randomized using a random seed based on the date and time of the generation. It produces an AVI video lasting 5 min 36 s.

Technical Validation

For the creation of audio scenes, two expert listeners evaluated the overall quality of the audio scenes following the ITU-R BS.775-4 recommendation¹⁴. The 3D audio scenes were designed to evoke specific emotional responses, guided by the Mexican validation of the International Affective Digitized Sounds (IADS-2) second edition¹⁵.

The presentation of audio scenes was performed in a previously calibrated virtual auditory environment, matching the acoustic parameters of the neuroengineering laboratory where the experimental procedure took place. The whole procedure is detailed in a previous work²³. Mainly, the auditory environment was calibrated in three parts: (1) acoustic measurements in the physical space, (2) settings adjustment of the virtual plugin to match the acoustic parameters of the physical room, and (3) psychoacoustic evaluation in terms of degradation by a sample of expert listeners.

Before the questionnaire application, the researchers explained each of the items in the STAI and SAM test, ensuring that the participants understood the words and messages of the items, resolving any of the concerns the participant may have before starting experimentation.

Lastly, behavioral and electrophysiological features of participants in resting state, during images presentation, and during auditory stimulation are shown in Fig. 5. The root mean square error (RMSE), computed between the raw and preprocessed electrophysiological signals, is shown in Fig. 6 for each condition and participant.

Fig. 5.

Distribution of responses for 60 participants presented in a jittered dot-box plot. The information is presented as follows: (a) State anxiety decile according to the STAI inventory, (b) Mean heart rate in beats per minute extracted from ECG recordings, (c) Mean conductance activity in μS from EDA recordings; and (d) valence, (e) arousal, and (f) dominance score from the SAM. The results are presented for three conditions with three different markers: (1) Resting state (square), (2) Images (circle), and (3) Audio (triangle).

Fig. 6.

RMSE across participants and experimental conditions. The heatmap shows the distribution of RMSE values computed for each participant (rows) and experimental condition (columns). Warmer colors indicate higher error magnitudes, reflecting greater deviation from the reference signal. Additionally, each cell contains the corresponding RMSE value.

In Fig. 5, it can be observed that the scores and electrophysiological features are within the typical ranges for each case. State anxiety decile ranges between 1 and 10, being low and high anxiety levels normalized according to age, sex and educational level. Heart rate ranges between 40 and 100 bpm, showing a centered distribution between 60–80 bpm, the normal heart rate according to the American College of Cardiology²⁴. Mean conductance is presented within the ranges specified by the g.GSRSENSOR acquisition system, having the distribution centered around 1 μS, value within the typical ranges for skin conductance response and skin conductance level²⁵. Valence, arousal and dominance scores range between 1 and 9, having the distribution centered at five, being the neutral point¹². Only dominance shows a skewness centered at seven, indicating that most participants had a sense of control over their own emotions.

In Fig. 6, elevated RMSE values in μV are observed in the ECG recordings for some participants (IDs 54, 55, 56, 57, 59, 64, 65, 69, 73, 76, 80, 94), indicating that these data should be filtered with appropriate preprocessing pipelines to remove noise and artifacts. In contrast, the EDA recordings show lower RMSE values in μS, suggesting that less noise was present in those signals. Only the recordings from one participant (ID 140) had the largest RMSE value.

Usage Notes

It is suggested to open EDF and SET/FDT files using EEGLAB toolbox in MATLAB. It is also recommended to revise the published articles related to STAI and SAM. The audio files with “26ch” indication are the 26-channel ambisonics mixes meant to be played on a 26p Lebedev loudspeaker array. The files with “binaural” term can be listened on headphones for 3D audio reproduction. The files with “room” word include DT990 pro headphones equalization and the acoustic parameters of the virtual neuroengineering laboratory, while the files with “dry” word are the anechoic versions without headphone calibration. For the use of the IAPS images, a request must be submitted at the website of the Centre for the Study of Emotions of the University of Florida. The full song EL cannot be redistributed due to copyright restrictions; the fragment is provided solely for research purposes to illustrate the experimental context. The song fragment may still be valid for emotional estimation purposes as 6-second musical fragments are traditionally used to elicit emotions¹⁵. For details on the preprocessing steps, please refer to the README files. Lastly, due to download size limit imposed by the repository, it is suggested to download the ZIP files separately, while the remaining files can be selected and downloaded together.

Author contributions

N.E.N.R. collected, revised, and organized the data presented in this work. He also produced the artwork and tables, wrote the original draft of the manuscript, and contributed to the conceptualization and methodology of data collection. S.C.S.M. collected and revised the data presented in this work and contributed to the original draft of the manuscript. She also produced the artwork of Fig. 3. R.G. and R.A.G.G. contributed to the generation of the audio files, revised the data, and contributed to the original draft of the manuscript. L.M.A.V. and D.I.I.Z. supervised the project, contributed to the conceptualization and methodology of data collection, provided funding for the research, and reviewed and edited the final version of the manuscript. A.A.C. and C.G.S.G. contributed to the participants’ health assessments, reviewed and edited the final version of the manuscript, and were responsible for the protocol submitted to the Ethics Committee of the School of Medicine of Tecnologico de Monterrey. All authors revised the final version of the manuscript.

Data availability

This dataset is available in the Research Data Hub repository of Tecnologico de Monterrey²² at 10.57687/FK2/LCB08J.

Code availability

The codes for video generation and electrophysiological recordings preprocessing are included in the repository with all the files described in this work.

Competing interests

The authors declare no competing interests.