Detection of Suicide Risk Using Vocal Characteristics: Systematic Review
Bibliographic record
Abstract
BACKGROUND: In an age when telehealth services are increasingly being used for forward triage, there is a need for accurate suicide risk detection. Vocal characteristics analyzed using artificial intelligence are now proving capable of detecting suicide risk with accuracies superior to traditional survey-based approaches, suggesting an efficient and economical approach to ensuring ongoing patient safety. OBJECTIVE: This systematic review aimed to identify which vocal characteristics perform best at differentiating between patients with an elevated risk of suicide in comparison with other cohorts and identify the methodological specifications of the systems used to derive each feature and the accuracies of classification that result. METHODS: A search of MEDLINE via Ovid, Scopus, Computers and Applied Science Complete, CADTH, Web of Science, ProQuest Dissertations and Theses A&I, Australian Policy Online, and Mednar was conducted between 1995 and 2020 and updated in 2021. The inclusion criteria were human participants with no language, age, or setting restrictions applied; randomized controlled studies, observational cohort studies, and theses; studies that used some measure of vocal quality; and individuals assessed as being at high risk of suicide compared with other individuals at lower risk using a validated measure of suicide risk. Risk of bias was assessed using the Risk of Bias in Non-randomized Studies tool. A random-effects model meta-analysis was used wherever mean measures of vocal quality were reported. RESULTS: The search yielded 1074 unique citations, of which 30 (2.79%) were screened via full text. A total of 21 studies involving 1734 participants met all inclusion criteria. Most studies (15/21, 71%) sourced participants via either the Vanderbilt II database of recordings (8/21, 38%) or the Silverman and Silverman perceptual study recording database (7/21, 33%). Candidate vocal characteristics that performed best at differentiating between high risk of suicide and comparison cohorts included timing patterns of speech (median accuracy 95%), power spectral density sub-bands (median accuracy 90.3%), and mel-frequency cepstral coefficients (median accuracy 80%). A random-effects meta-analysis was used to compare 22 characteristics nested within 14% (3/21) of the studies, which demonstrated significant standardized mean differences for frequencies within the first and second formants (standardized mean difference ranged between -1.07 and -2.56) and jitter values (standardized mean difference=1.47). In 43% (9/21) of the studies, risk of bias was assessed as moderate, whereas in the remaining studies (12/21, 57%), the risk of bias was assessed as high. CONCLUSIONS: Although several key methodological issues prevailed among the studies reviewed, there is promise in the use of vocal characteristics to detect elevations in suicide risk, particularly in novel settings such as telehealth or conversational agents. TRIAL REGISTRATION: PROSPERO International Prospective Register of Systematic Reviews CRD420200167413; https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42020167413.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.010 | 0.063 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.010 | 0.010 |
| Bibliometrics | 0.013 | 0.013 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.005 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".