[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-trim-audio-deepfake-detection-errors-with-tiny-add-on":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},6867,"researchers-trim-audio-deepfake-detection-errors-with-tiny-add-on","Researchers Trim Audio Deepfake Detection Errors With Tiny Add-On","A new refinement technique cuts audio deepfake detection errors by training just 10 million extra parameters atop a frozen 598 million parameter model.","A new technique lets an already-trained audio deepfake detector catch more fakes without any new training data or a rebuilt model.\n\nResearchers call the method CoReLoop. It takes an existing SSL-based detector - a large neural network trained via self-supervised learning - and adds a loop: the model reprocesses its own encoder outputs, refining its verdict on a second pass. Naively feeding a detector's own outputs back into itself makes accuracy worse, so the team added lightweight adapter modules that control how the loop updates and keep the refined answer aligned with the original classifier. Only about 10 million of the full 598 million parameters are trained, and everything runs on top of the detector's original data. Across 14 cross-domain test sets, two passes cut the pooled equal error rate from 4.85% to 3.74%. An optional halting head, which decides per-clip whether a second pass is worth it, gets a similar 3.73% EER while averaging just 1.18 passes per utterance.\n\nThe real problem in deepfake detection isn't building an accurate model today - it's that tomorrow's fake-audio generator wasn't in the training set. Collecting data for every future attack is impractical, so most detectors quietly degrade the moment a new synthesis method shows up. CoReLoop sidesteps that by squeezing more signal out of a detector that already exists, rather than promising a model that generalizes perfectly out of the box.\n\nCall it a tune-up, not a rewrite. A one-point EER drop is real but modest, and the 14 test sets are still academic benchmarks - the attacks circulating on social platforms and robocalls will be the harder test.","[\"deepfake-detection\",\"audio-ai\",\"ai-research\",\"security\"]","2026-09-18T04:00:00.000Z","2026-09-18T20:10:36.420Z","2026-09-18T20:10:48.328Z","published",null,[],"security",[26,27,28,24],"deepfake-detection","audio-ai","ai-research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.19818",0,{"sections":35},[36,40,43,48,53,57,61,66,70,75,80,85,90,95],{"name":37,"slug":38,"count":39,"latest_published_at":18},"AI","ai",4031,{"name":41,"slug":24,"count":42,"latest_published_at":18},"Security",654,{"name":44,"slug":45,"count":46,"latest_published_at":47},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":18},"Hardware","hardware",155,{"name":58,"slug":59,"count":60,"latest_published_at":18},"Science","science",121,{"name":62,"slug":63,"count":64,"latest_published_at":65},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":67,"slug":68,"count":69,"latest_published_at":18},"Dev Tools","dev-tools",78,{"name":71,"slug":72,"count":73,"latest_published_at":74},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":86,"slug":87,"count":88,"latest_published_at":89},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":91,"slug":92,"count":93,"latest_published_at":94},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":96,"slug":97,"count":98,"latest_published_at":99},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]