Abstract
Abstract
Introduction
There are many barriers to mental health services, including cost and stigma. Even when individuals receive professional care, assessments are intermittent and may be limited in part by the cyclical nature of psychiatric symptoms. The human voice might have the potential to serve as a valuable biomarker in the identification, early diagnosis or monitoring of psychiatric conditions. Therefore, this protocol presents a proposed scoping review with the aim of synthesising existing knowledge on the application of artificial intelligence (AI) or machine learning (ML) in the management of individuals at risk of/suffering from schizophrenia through audio samples as a biomarker.
Methods and analysis
Guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews guidelines and Arksey & O’Malley’s scoping review framework (with recent advancements), we systematically mapped the literature on the application of voice-based biomarkers in schizophrenia. Several databases (PubMed/MEDLINE, Scopus, Web of Science, IEEE Xplore, Embase, Compendex, CINAHL, Scientific Information Database, Magiran, IranMedex and Barakat knowledge network system) will be systematically searched for relevant studies through 2025. All searches will be conducted for peer-reviewed articles/studies published in Persian and English between 1 January 2012 and 1 September 2025. Two researchers will independently carry out screening of the included studies and extraction of data. Any discrepancies will be resolved by consensus. In case no initial consensus is reached, a third researcher will be consulted to make a decision. Findings will be presented narratively in the form of text, summary tables, charts and figures for each research question.
Ethics and dissemination
This proposed scoping review is based on publicly available information and is also a review of primary studies, so ethics and publication ethics approval are not required because all data from this study have been previously published. The findings of this review will be published in a peer-reviewed journal and presented at national or international congresses and conferences. Importantly, the initial results from this review will serve as a basis for the design and validation of an intelligent clinical decision support system based on acoustic biomarkers for patients with schizophrenia, using AI or ML techniques.
Systematic review registration
Not registered.
Keywords: Adult psychiatry, PSYCHIATRY, eHealth, Information technology, Health informatics, Digital Technology
STRENGTHS AND LIMITATIONS OF THIS STUDY.
Developed a comprehensive search protocol with a medical librarian to ensure extensive database coverage.
Implemented dual independent screening and data extraction to reduce selection bias.
Excluded non-English and non-Persian studies due to practical constraints related to translation and quality assurance.
Designed the study protocol with input from a multidisciplinary team of clinicians, artificial intelligence experts and scoping review specialists.
Introduction
Schizophrenia is a complex mental disorder that affected 23.6 million people worldwide by 2019.1 Schizophrenia is characterised by symptoms such as hallucinations, delusions, disorganised thinking and impaired social functioning.2 These symptoms can vary widely among individuals and often fluctuate in severity over time.3 Traditionally, schizophrenia diagnosis and management have relied on clinical interviews, self-reports and observational assessments based on clinicians' judgement. While these methods are necessary for diagnosing and managing the disease, they are subjective and might fail to capture subtle changes in the patient’s condition.4
In recent years, advancements in artificial intelligence (AI) and machine learning (ML) have offered new avenues for more precise and scalable diagnostic tools. A particularly promising avenue is the use of vocal characteristics as digital biomarkers. Subtle features and patterns in an individual’s speech are collected and analysed by digital devices.5 These features include speech rate, pitch, tone, rhythm and various acoustic and linguistic characteristics.6 7 Studies have shown that individuals with schizophrenia often exhibit distinctive vocal patterns that can be analysed using AI-based voice biomarkers to address health-related questions similar to traditional biomarkers.8,10
The use of AI to analyse vocal biomarkers in schizophrenia has several potential advantages; this method is non-invasive and cost-effective for early diagnosis. By analysing speech samples, AI algorithms can detect anomalies that may appear before the onset of overt clinical symptoms, allowing for earlier intervention and prevention of disease progression.11 This is particularly valuable given the importance of early detection in improving long-term outcomes for individuals with schizophrenia. Furthermore, AI-based voice analysis can enhance continuous monitoring and disease management. Patients can regularly provide speech samples through simple voice recordings, which can be analysed to track changes in their condition over time. This enables personalised treatment plans and timely adjustments in medication or therapeutic interventions, reducing the burden on healthcare providers by automating parts of the assessment process and allowing them to focus on more complex aspects of patient care.12
Another advantage of using AI to analyse vocal biomarkers is the improvement in diagnostic accuracy and reliability.13 Common vocal/speech pathologies in schizophrenia (eg, reduced prosody, monotone speech, irregular pitch or pauses) can be analysed by AI.8 While clinicians are mostly attracted to the process of thought reflected in speech of patients, these tools might detect subtle patterns potentially missed by clinicians through large datasets and advanced algorithms,14 improving diagnostic accuracy, reducing errors and facilitating appropriate treatment.15 The potential impact of this approach extends beyond individual patients. Data collected from vocal biomarker analysis can contribute to a better understanding of the underlying mechanisms of schizophrenia, contributing to new insights and advancements in therapy.16 Additionally, the scalability of AI-based voice analysis means it can be applied on a large scale, benefiting a broader population and improving public health outcomes.
To the best of our knowledge, previous systematic reviews have examined distinctive acoustic patterns in schizophrenia and their correlation with clinical features.8 Another systematic review has addressed the use of speech for automated assessment across a broader range of psychiatric disorders.12 However, in this scoping review, a conceptual framework will be presented to demonstrate the applications of AI and ML in analysing the voices of individuals with or at risk of schizophrenia as a biomarker. This study aims to synthesise the existing knowledge on the application of AI and ML in managing individuals with or at risk of schizophrenia through voice samples as a biomarker. The present study will describe the data collection methodology, AI algorithms used and potential impact on clinical outcomes. Leveraging AI’s capabilities may significantly enhance current understanding and management of schizophrenia, with potential to improve patient outcomes and quality of life.
Methods and analysis
Review method
The summary of this protocol has been reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Protocols statement.17 18(see online supplemental additional file 1). The proposed scoping review will be reported by the reporting guidance provided in the PRISMA extension for Scoping Reviews (PRISMA-ScR).19 This protocol could not be registered in PROSPERO as it is a scoping review, which falls outside the scope of that registry. This study was designed based on the methodological framework developed by Arksey and O’Malley,20 the methodology described in the Joanna Briggs Institute (JBI) Reviewers’ Manual, which other authors have further refined.21 22 A scoping review is designed to examine the extent, scope and nature of research activity on a topic of interest and to identify research gaps in the available literature.23 Therefore, this review will be useful in identifying and mapping the existing findings regarding the application of AI based on voice biomarkers in the management of people at risk of/suffering from schizophrenia disorders. The framework comprises six sequential stages for conducting a rigorous scoping review: (1) identifying the research question, (2) identifying relevant studies, (3) selecting studies, (4) charting the data, (5) collating, summarising and reporting the results and (6) conducting a consultation exercise.20 The planned study timeline encompasses all six phases of the Arksey-O’Malley framework and is scheduled from September 2025 to March 2026 to guide the systematic implementation and completion of the review.
Stage 1: identifying the research question
The main research question for this study is: ‘How is voice used as a digital biomarker in clinical applications through AI or ML techniques for managing patients with schizophrenia? What are the applications of AI and ML in managing individuals with or at risk of schizophrenia through voice samples as a digital biomarker? The specific subquestions include:
-
What are the clinical applications of voice biomarker analysis in schizophrenia management, including both assessment (diagnosis, symptom monitoring) and treatment (therapeutic intervention evaluation)? What are the standard methodologies for voice recording in schizophrenia voice biomarker studies, including:
Tasks: free speech, reading passages, sustained vowel phonation, or structured interviews.
File formats: lossless (WAV, FLAC) versus compressed (MP3, AAC) formats.
Equipment: clinical-grade microphones versus consumer devices (smartphones, tablets)?
What are the audio features used as input and output data based on AI and ML techniques for managing patients with schizophrenia?
Which AI/ML models (eg, SVM (Support Vector Machine), Neural Networks, Random Forest) show the highest efficacy in analysing voice biomarkers for schizophrenia diagnosis and monitoring?
What standardised metrics (eg, accuracy, sensitivity/specificity, AUC (Area Under the Curve)-ROC (Receiver Operating Characteristic), F1-score, computational efficiency) are used to evaluate the performance of AI models in voice-based schizophrenia detection?
To what extent do current voice biomarker datasets and AI models for schizophrenia account for demographic variability (eg, age, sex, ethnicity, linguistic background) and how does this variability impact model performance and clinical generalisability?
What barriers (eg, dataset diversity) and facilitators (eg, clinician acceptance) influence implementation?
Inclusion and exclusion criteria
All eligible studies, without time restrictions and limited to English and Persian languages, will be included in the analysis.
Inclusion criteria
Articles published in peer-reviewed journals
Primary studies written in English and Persian.
Research published no earlier than 2012.
Studies that have used voice as input data.
Studies that have used at least one AI or ML algorithm.
Research aiming at the application of AI and ML in managing individuals with or at risk of schizophrenia through voice samples as a biomarker, excluding those related to respiratory, gastrointestinal and other disorders affecting voice.
Exclusion criteria
Studies published without peer review.
Research written in languages other than English and Persian.
Studies published before 2012.
Studies that have not directly used voice data, meaning research that has used various non-verbal forms of input data, such as written transcriptions, digital images, videos, electroencephalogram and signals produced during phonation.
Studies that classify voice-affecting diseases without using a ML or AI approach.
Studies that classify vocal disorders related to respiratory, gastrointestinal and other conditions affecting voice.
Stage 2: identification of relevant studies
A search strategy will be formulated using the population, intervention, comparison and outcome (PICO) framework. The commonly used terms, suggested by the authors for finding relevant articles, will include AI, voice and schizophrenia. Following the initial search, new relevant keywords will be discovered. Subsequently, a comprehensive search strategy will be created by adding these new keywords and will be finalised and approved by all members of the research team. A systematic search will be conducted across eleven databases (PubMed/MEDLINE, Scopus, Web of Science, IEEE Xplore, Embase, Compendex, CINAHL, Scientific Information Database, Magiran, IranMedex and Barakat Knowledge Network System) to identify relevant studies published up to and including 1 September 2025. Additionally, the Google Scholar and ClinicalTrials.gov databases will be searched to uncover grey literature. The core search strategy developed for PubMed/MEDLINE will be systematically adapted to each additional database (eg, EMBASE, PsycINFO, IEEE Xplore) by modifying syntax as needed while preserving the original conceptual structure and Boolean logic (see online supplemental additional file 2).
To enhance transparency in our grey literature search, we will:
Systematically document searches of conference proceedings (eg, Schizophrenia International Research Society meetings), preprint servers (bioRxiv, medRxiv) and clinical trial registries (ClinicalTrials.gov) using the same structured approach as our database searches.
Explicitly report the date ranges, platforms searched and search strings employed.
Acknowledge limitations in replicability due to the dynamic nature of grey literature and platform-specific search functionalities. The PICO framework will include the following components, although the ‘comparison’ component is not applicable due to the nature of this review:
Population: patients with or at risk of schizophrenia.
Intervention: The use of AI and ML techniques to manage patients with schizophrenia through voice samples. Management includes assessment (diagnosis, monitoring) and treatment (therapeutic intervention evaluation).
-
Outcome:
Performance metrics: accuracy, precision, sensitivity (recall), specificity, AUC, F1-score and other reported model evaluation measures.
Clinical utility: where available, clinical correlation measures (eg, symptom severity prediction).
The search strategy will primarily be based on Medical Subject Headings terms and will be adjusted to meet the specific search requirements of each database. Boolean logic and proximity operators (AND/OR) will be used to combine and refine search terms and keywords. Filter options will be set to retrieve articles published from 1 January 2012 to 1 January 2025. This period has been selected in consultation with experts in the fields of medicine and health information technology, considering the development of new technologies for healthcare. The year 2012 was selected as the cut-off for this review because AI-based voice analysis in schizophrenia gained significant prominence thereafter, with advancements in ML and AI methodologies. This ensures the inclusion of contemporary techniques and relevant studies that reflect the current state of the field.
Stage 3: study selection
The study selection will be conducted following the PRISMA-ScR flowchart. EndNote V.8 software will be used for reference management, removal of duplicates and the screening process. The screening process will involve three stages. First, titles and abstracts retrieved from the initial searches will be independently screened by two researchers simultaneously based on the mentioned eligibility criteria. Second, the full text of the potentially relevant evidence will be reviewed and assessed in detail by the same researchers according to the inclusion and exclusion criteria. Third, to identify further relevant publications not found in the databases, the reference lists and citations of all included studies will be manually reviewed. The reasons for excluding studies at the full-text level will be recorded and noted in the PRISMA flowchart. In rare cases where studies are eligible but have incomplete or limited information, such as only having the abstract available, the authors will be contacted via email to obtain additional details and the full text. Any potential disagreements will be resolved through discussion between the two researchers. If no agreement is reached, a third researcher will review the study.
Stage 4: data charting
To ensure comprehensive and relevant data extraction, data extraction tables will be designed in MS Word 2016. The elements of the extraction chart will be based on the guidance and instructions from the JBI extraction tool. The data of interest will include: author(s), year of publication, research objectives/question(s), study type, general study information, study characteristics, participant characteristics, type of voice data recorded, characteristics of input and output voice data, types of ML or AI models and criteria used to evaluate the performance and efficiency of ML and AI models (see online supplemental additional file 3). Where available, data on accuracy, sensitivity and specificity for each model will be collected. Initially, data extraction will be carried out by two members of the research team (MAB and AL) for 20% of the studies included in the review. If no disagreements arise, they will continue to collaboratively extract data for the remaining studies. Other researchers (SS and SF) will critically review the data extraction process to assess the accuracy of the extracted data.
Stage 5: data analysis and reporting the results
To summarise temporal trends and capture heterogeneity, studies will be grouped into summary tables under the title, ‘Application of AI and ML in Managing Individuals with or at Risk of Schizophrenia through Voice Samples as a Biomarker’. All studies that successfully meet the inclusion and exclusion criteria will be eligible for data analysis. The findings will be presented in text, summary tables, charts and figures in a section dedicated to each research question.
Stage 6: consultation exercise
The initial findings will serve as the basis for the design and validation of an intelligent clinical decision support system based on vocal biomarkers for patients with schizophrenia, using AI or ML techniques.
Ethics and dissemination
Ethics
This scoping review protocol received ethical approval from the Research Ethics Committee of Tabriz University of Medical Sciences. Although the study dose not involve patient data or clinical procedures, this approval was sought to ensure full adherence to ethical standards for review methodologies.
Dissemination
We will seek to publish the findings of this scoping review in a peer-reviewed journal in the fields of psychiatry, digital health or biomedical informatics. The findings will also be presented at relevant national and international conferences. The results of this study will be presented in the form of a seminar or journal club for researchers at the Development and Technology Center, Tabriz University of Medical Sciences, Tabriz, Iran, as well as the Tabriz Psychiatric Research Center.
Discussion
AI-based biomarkers hold considerable promise for enhancing healthcare delivery. AI tools use digital biomarkers to support clinical decision-making and advance evidence-based care, ultimately contributing to personalised medicine. This approach has the potential to improve health outcomes, minimise unjustified variations in clinical practice and reduce the economic burden on patients and healthcare systems.24
The human voice has complex yet unique acoustic signatures that vary depending on the harmony between breathing, intonation, articulation and prosody of the individual. Recent technological advances, especially in AI and voice analysis, have positioned voice as a cost-effective, non-invasive and accessible digital biomarker for diagnosing various pathologies. The primary aim of this protocol is to map the current use of voice combined with AI models in managing patients with schizophrenia. Hence, this protocol will highlight the current and potential applications of AI in the analysis of the voice of schizophrenic patients as a clinically relevant digital biomarker in mental health.
The proposed scoping review has several strengths. The review will follow rigorous methodological frameworks to ensure transparency, reproducibility and comprehensiveness. We will employ a comprehensive search strategy using multiple academic databases and grey literature sources. The evidence identified and synthesised in this study will help to outline gaps in the evidence for AI-based voice biomarker implementation, what has worked and what needs to be improved for effective implementation. This scoping review will also provide an evidence base for future research programmes towards implementing an AI-based voice biomarker-based clinical decision support system for schizophrenia patients using AI or ML techniques.
Nevertheless, a key limitation of this study is the exclusion of non-English and non-Persian language publications due to resource constraints related to translation and quality assurance, which may limit the breadth of evidence captured. Additional methodological limitations include the potential for missing unpublished or grey literature data, variability in study designs and AI methodologies and heterogeneity in acoustic biomarkers used across studies, all of which may pose challenges for data synthesis and generalisability of findings.
Supplementary material
Acknowledgements
The authors would like to acknowledge the Student Research Committee of Tabriz University of Medical Sciences for its financial support.
Footnotes
Funding: The research protocol was approved and supported by the Student Research Committee, Tabriz University of Medical Sciences (grant number: 76150). The funding body had no role in making the decision to design the study; collection, analysis and interpretation of data; nor in writing the manuscript. The funder only provided funding for conducting the study.
Prepublication history and additional supplemental material for this paper are available online. To view these files, please visit the journal online (https://doi.org/10.1136/bmjopen-2025-099475).
Provenance and peer review: Not commissioned; externally peer reviewed.
Patient consent for publication: Not applicable.
Patient and public involvement: Patients and/or the public were not involved in the design, conduct, reporting or dissemination plans of this research.
References
- 1.Solmi M, Seitidis G, Mavridis D, et al. Incidence, prevalence, and global burden of schizophrenia - data, with critical appraisal, from the Global Burden of Disease (GBD) 2019. Mol Psychiatry. 2023;28:5319–27. doi: 10.1038/s41380-023-02138-4. [DOI] [PubMed] [Google Scholar]
- 2.Jauhar S, Johnstone M, McKenna PJ. Schizophrenia. Lancet. 2022;399:473–86. doi: 10.1016/S0140-6736(21)01730-X. [DOI] [PubMed] [Google Scholar]
- 3.Fatani BZ, Aldawod RA, Alhawaj FA. Schizophrenia : Etiology, Pathophysiology and Management : A Review. Egyptian Journal of Hospital Medicine . 2017;69:2640–6. doi: 10.12816/0042241. [DOI] [Google Scholar]
- 4.Orsolini L, Pompili S, Volpe U. Schizophrenia: A Narrative Review of Etiopathogenetic, Diagnostic and Treatment Aspects. J Clin Med. 2022;11:5040. doi: 10.3390/jcm11175040. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Fagherazzi G, Fischer A, Ismael M, et al. Voice for Health: The Use of Vocal Biomarkers from Research to Clinical Practice. Digit Biomark. 2021;5:78–88. doi: 10.1159/000515346. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Babrak LM, Menetski J, Rebhan M, et al. Traditional and Digital Biomarkers: Two Worlds Apart? Digit Biomark. 2019;3:92–102. doi: 10.1159/000502000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Coravos A, Khozin S, Mandl KD. Developing and adopting safe and effective digital biomarkers to improve patient outcomes. NPJ Digit Med. 2019;2:14. doi: 10.1038/s41746-019-0090-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Parola A, Simonsen A, Bliksted V, et al. Voice patterns in schizophrenia: A systematic review and Bayesian meta-analysis. Schizophr Res. 2020;216:24–40. doi: 10.1016/j.schres.2019.11.031. [DOI] [PubMed] [Google Scholar]
- 9.Zhao Q, Wang W-Q, Fan H-Z, et al. Vocal acoustic features may be objective biomarkers of negative symptoms in schizophrenia: A cross-sectional study. Schizophr Res. 2022;250:180–5. doi: 10.1016/j.schres.2022.11.013. [DOI] [PubMed] [Google Scholar]
- 10.Dorsey ER, Papapetropoulos S, Xiong M, et al. The First Frontier: Digital Biomarkers for Neurodegenerative Disorders. Digit Biomark. 2017;1:6–13. doi: 10.1159/000477383. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Khare M, Acharya S, Shukla S, et al. Utilising Artificial Intelligence (AI) in the Diagnosis of Psychiatric Disorders: A Narrative Review. JCDR . 2024;18 doi: 10.7860/JCDR/2023/61698.19249. [DOI] [Google Scholar]
- 12.Idrisoglu A, Dallora AL, Anderberg P, et al. Applied Machine Learning Techniques to Diagnose Voice-Affecting Conditions and Disorders: Systematic Literature Review. J Med Internet Res. 2023;25:e46105. doi: 10.2196/46105. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Bianco MJ, Gerstoft P, Traer J, et al. Machine learning in acoustics: Theory and applications. J Acoust Soc Am. 2019;146:3590–628. doi: 10.1121/1.5133944. [DOI] [PubMed] [Google Scholar]
- 14.Low DM, Bentley KH, Ghosh SS. Automated assessment of psychiatric disorders using speech: A systematic review. Laryngoscope Investig Otolaryngol. 2020;5:96–116. doi: 10.1002/lio2.354. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Corcoran CM, Carrillo F, Fernández-Slezak D, et al. Prediction of psychosis across protocols and risk cohorts using automated language analysis. World Psychiatry. 2018;17:67–75. doi: 10.1002/wps.20491. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Parola A, Trenckner Jessen E, Rybner A, et al. Vocal markers of schizophrenia: assessing the generalizability of machine learning models and their clinical applicability. Psychiatry and Clinical Psychology. 2023 doi: 10.1101/2024.11.06.24316839. Preprint. [DOI] [PubMed]
- 17.Shamseer L, Moher D, Clarke M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015: elaboration and explanation. BMJ. 2015;350:g7647. doi: 10.1136/bmj.g7647. [DOI] [PubMed] [Google Scholar]
- 18.Moher D, Shamseer L, Clarke M, et al. Preferred reporting items for systematic review and meta-analysis protocols (PRISMA-P) 2015 statement. Syst Rev. 2015;4:1. doi: 10.1186/2046-4053-4-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Tricco AC, Lillie E, Zarin W, et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann Intern Med. 2018;169:467–73. doi: 10.7326/M18-0850. [DOI] [PubMed] [Google Scholar]
- 20.Arksey H, O’Malley L. Scoping studies: towards a methodological framework. Int J Soc Res Methodol. 2005;8:19–32. doi: 10.1080/1364557032000119616. [DOI] [Google Scholar]
- 21.Levac D, Colquhoun H, O’Brien KK. Scoping studies: advancing the methodology. Implement Sci. 2010;5:69. doi: 10.1186/1748-5908-5-69. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Colquhoun HL, Levac D, O’Brien KK, et al. Scoping reviews: time for clarity in definition, methods, and reporting. J Clin Epidemiol. 2014;67:1291–4. doi: 10.1016/j.jclinepi.2014.03.013. [DOI] [PubMed] [Google Scholar]
- 23.Peters MDJ, Godfrey CM, Khalil H, et al. Guidance for conducting systematic scoping reviews. Int J Evid Based Healthc. 2015;13:141–6. doi: 10.1097/XEB.0000000000000050. [DOI] [PubMed] [Google Scholar]
- 24.Arya SS, Dias SB, Jelinek HF, et al. The convergence of traditional and digital biomarkers through AI-assisted biosensing: A new era in translational diagnostics? Biosens Bioelectron. 2023;235:115387. doi: 10.1016/j.bios.2023.115387. [DOI] [PubMed] [Google Scholar]