[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-speech-model-skips-engineered-targets-still-tops-benchmarks":10,"sections":35},{"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":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},8652,"new-speech-model-skips-engineered-targets-still-tops-benchmarks","New Speech Model Skips Engineered Targets, Still Tops Benchmarks","A 57M-parameter speech model skips engineered targets and still outperforms rivals on transcription benchmarks.","A new speech AI model called GLaS-JEPA gets state-of-the-art results by predicting its own representations instead of relying on hand-engineered targets.\n\nResearchers built GLaS-JEPA, a self-supervised speech learning framework that directly predicts the current encoder's continuous representations at masked positions in the audio. That breaks from prior methods, which lean on contrastive learning, discrete targets, or a separate exponential-moving-average target encoder to keep the model from collapsing into trivial solutions. Instead, GLaS-JEPA uses a technique called SIGReg to regularize the representation space directly and prevent that collapse. The team pretrained a 57-million-parameter version of the model on 960 hours of the LibriSpeech dataset.\n\nOn frozen-encoder benchmarks from the SUPERB suite, the model scored a 6.89% word error rate on speech recognition and a 25.87% character error rate on slot filling, beating the best non-distilled models under 90 million parameters by 43.1% and 22.0% respectively. That matters because speech pretraining recipes have gotten increasingly elaborate, stacking extra machinery just to stop representations from collapsing. If a simpler regularization step can match or beat that complexity, it strips real engineering overhead out of building speech foundation models.\n\nStill, LibriSpeech is a clean, English-only benchmark, and 57M parameters is modest by current standards - whether SIGReg holds up at larger scale or on messier, multilingual audio is still an open question.","[\"speech-recognition\",\"self-supervised-learning\",\"machine-learning\",\"ai-research\"]","2026-09-30T04:00:00.000Z","2026-09-30T17:18:10.453Z","2026-09-30T17:18:15.292Z","published",null,[],"ai",[26,27,28,29],"speech-recognition","self-supervised-learning","machine-learning","ai-research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.37798",0,{"sections":36},[37,40,44,48,53,58,62,67,72,76,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5181,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",791,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",417,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",284,"2026-09-29T21:00:00.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",194,"2026-09-29T13:16:04.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",155,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",142,"2026-09-29T18:38:03.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Software","software",91,"2026-09-25T20:55:00.000Z",{"name":73,"slug":74,"count":75,"latest_published_at":18},"Dev Tools","dev-tools",90,{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",83,"2026-09-29T21:51:36.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"General","general",49,"2026-09-28T16:44:57.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"Gaming","gaming",48,"2026-09-25T18:35:21.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]