{"id":"W4408142317","doi":"10.1016/j.jvoice.2025.02.020","title":"Reliability and Task Effects in CAPE-V Auditory-Perceptual Voice Assessments: Insights From the PVQD30 Subset","year":2025,"lang":"en","type":"article","venue":"Journal of Voice","topic":"Voice and Speech Disorders","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Reliability (semiconductor); Psychology; Perception; Task (project management); Audiology; Speech recognition; Cognitive psychology; Computer science; Neuroscience; Medicine; Engineering","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005173712,0.0001701588,0.000450489,0.0001413663,0.00008963518,0.00004050348,0.0002037063,0.0001418765,0.00002493348],"category_scores_gemma":[0.001065377,0.0001061359,0.0001143601,0.0003248858,0.0001162382,0.0002663723,0.00006082295,0.0007752648,0.00001923541],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001535897,"about_ca_system_score_gemma":0.0002941997,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000438005,"about_ca_topic_score_gemma":0.0006031877,"domain_scores_codex":[0.9984055,0.0002652952,0.0005177573,0.0002116794,0.0004017769,0.0001979618],"domain_scores_gemma":[0.9979469,0.001204152,0.0002132426,0.0003124263,0.0002086143,0.0001146875],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002681568,0.00258412,0.6864672,0.0008296184,0.001112614,0.0009029832,0.01159744,0.0001439874,0.07655701,0.0002956238,0.1951123,0.02171557],"study_design_scores_gemma":[0.003752653,0.0004075629,0.9458689,0.0006104317,0.0002851962,0.00002652624,0.002266947,0.0001680771,0.0002996531,0.001255708,0.04494367,0.0001146769],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9905847,0.002393611,0.0002451723,0.004290619,0.0009003922,0.0002790961,0.000006032375,0.00001054564,0.001289875],"genre_scores_gemma":[0.9952346,0.0003866142,0.0005143263,0.003109832,0.0004409274,0.000004992061,0.000005611333,0.0000115047,0.0002915947],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2594017,"threshold_uncertainty_score":0.4328095,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00883875545002901,"score_gpt":0.2960607339214456,"score_spread":0.2872219784714166,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}