{"id":"W2795552730","doi":"10.1177/0023830918765012","title":"Investigating Perceptual Biases, Data Reliability, and Data Discovery in a Methodology for Collecting Speech Errors From Audio Recordings","year":2018,"lang":"en","type":"article","venue":"Language and Speech","topic":"Speech Recognition and Synthesis","field":"Computer Science","cited_by":30,"is_retracted":false,"has_abstract":true,"ca_institutions":"Simon Fraser University","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Computer science; Speech recognition; Reliability (semiconductor); Perception; Sound quality; Sound recording and reproduction; Natural language processing; Psychology; Acoustics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.466764,0.001677935,0.001846413,0.005316838,0.005698851,0.009688697,0.005273869,0.003968263,0.001076906],"category_scores_gemma":[0.7582012,0.003229382,0.001924267,0.006023533,0.01449431,0.008271872,0.009393575,0.004119995,0.000621506],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004291526,"about_ca_system_score_gemma":0.009669742,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004292661,"about_ca_topic_score_gemma":0.005329422,"domain_scores_codex":[0.3458895,0.509827,0.05000615,0.0338039,0.05812817,0.002345173],"domain_scores_gemma":[0.110797,0.6690948,0.06136304,0.1121837,0.04501507,0.001546365],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004435875,0.001628043,0.3651745,0.007296885,0.003439881,0.002551274,0.1770671,0.0213576,0.04526838,0.07642582,0.004456319,0.2908984],"study_design_scores_gemma":[0.002526935,0.009013207,0.301231,0.006166672,0.002653036,0.008907728,0.04392903,0.1373875,0.1162374,0.2840826,0.08564736,0.002217493],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1009547,0.0005286444,0.8886588,0.001730062,0.0002035767,0.004841433,0.00037436,0.0003313679,0.002377073],"genre_scores_gemma":[0.302174,0.0002265736,0.6854566,0.000949218,0.000216693,0.00996147,0.0003659338,0.0002419868,0.0004074865],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.533236,"threshold_uncertainty_score":0.6575749,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.231578198380963,"score_gpt":0.3871973809097454,"score_spread":0.1556191825287824,"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."}}