{"id":"W4385422383","doi":"10.31234/osf.io/h9mna","title":"Leveraging Natural Language Processing Models to Automate Speech-Intelligibility Scoring","year":2023,"lang":"en","type":"preprint","venue":"","topic":"Noise Effects and Management","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Baycrest Hospital; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Intelligibility (philosophy); Active listening; Computer science; Speech recognition; Speech perception; Perception; Natural language processing; Artificial intelligence; Audiology; Psychology; Communication; Medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001992011,0.0004618794,0.0006403359,0.0003556385,0.0006064599,0.0001107359,0.0005806527,0.0002953843,0.000189912],"category_scores_gemma":[0.0001623256,0.0003872351,0.000158809,0.0002881002,0.00002387707,0.0001694108,0.004282846,0.001897823,0.001109777],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006235474,"about_ca_system_score_gemma":0.0003293073,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005051828,"about_ca_topic_score_gemma":0.0008293137,"domain_scores_codex":[0.9962527,0.0003158834,0.000783461,0.001147121,0.0005042692,0.0009966128],"domain_scores_gemma":[0.9980537,0.0002634138,0.0002679481,0.001001153,0.000177075,0.0002367248],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002951254,0.0002979437,0.003328739,0.05257917,0.0005526528,0.0007532205,0.1445895,0.08686695,0.002490261,0.004749511,0.08778404,0.6157129],"study_design_scores_gemma":[0.0009627885,0.00006091188,0.006929182,0.01485095,0.0002105664,0.000002380175,0.02643124,0.9236624,0.00114856,0.01967406,0.003796335,0.002270633],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8224527,0.0009863203,0.08480277,0.004152976,0.0121026,0.008897739,0.00003686759,0.006915982,0.05965201],"genre_scores_gemma":[0.9400319,0.00002580864,0.01071077,0.001848211,0.000749352,0.0005170103,0.00005740193,0.000126946,0.04593262],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8367954,"threshold_uncertainty_score":0.999858,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1695954069559005,"score_gpt":0.4634092932595349,"score_spread":0.2938138863036345,"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."}}