{"id":"W4416397568","doi":"10.1016/j.specom.2025.103330","title":"Towards unsupervised speech recognition without pronunciation models","year":2025,"lang":"en","type":"article","venue":"Speech Communication","topic":"Speech Recognition and Synthesis","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Institute for Information and Communications Technology Promotion","keywords":"Pronunciation; Word (group theory); Pipeline (software); Unsupervised learning; Segmentation; Vocabulary; Word error rate; Speech corpus; Joint (building)","routes":{"ca_aff":true,"ca_fund":false,"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":[],"consensus_categories":[],"category_scores_codex":[0.001010655,0.0002089547,0.000244247,0.0003759978,0.0003289724,0.000308789,0.001486378,0.000174352,0.0000706932],"category_scores_gemma":[0.0002463282,0.0002169405,0.0001124885,0.0009634183,0.00005831788,0.001447841,0.0003700171,0.000274669,0.0002772496],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002331193,"about_ca_system_score_gemma":0.0002220518,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001970857,"about_ca_topic_score_gemma":0.0001096722,"domain_scores_codex":[0.9980234,0.0003807285,0.0004606577,0.0004295295,0.0004282964,0.0002774136],"domain_scores_gemma":[0.9973208,0.0001474349,0.000173776,0.001698576,0.000582809,0.00007664727],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0000142374,0.0001310929,0.0000864354,0.00002091754,0.00002793775,0.000001344237,0.0002338953,0.000009280895,0.0007623464,0.01295717,0.001247294,0.984508],"study_design_scores_gemma":[0.001487086,0.00007388677,0.002406129,0.0004074087,0.00006992715,0.0000346124,0.0002605683,0.260279,0.176331,0.5465824,0.01136145,0.0007065627],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05130233,0.0005846073,0.6044388,0.01438463,0.0004713995,0.001317414,0.00001636034,0.001270779,0.3262137],"genre_scores_gemma":[0.405525,0.0005732451,0.5906923,0.001717726,0.0000380643,0.0001630266,0.0001344164,0.00001907629,0.001137105],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9838015,"threshold_uncertainty_score":0.8846576,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0460206974990346,"score_gpt":0.2815052824270141,"score_spread":0.2354845849279795,"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."}}