[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-new-reward-trick-speeds-up-live-speech-transcription":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},6614,"new-reward-trick-speeds-up-live-speech-transcription","New Reward Trick Speeds Up Live Speech Transcription","A new training method rewards speech recognition models for both accuracy and timing, cutting error rates by over 30 percent at the lowest latency setting.","A new training trick makes live speech-to-text systems less trigger-shy about locking in words early.\n\nThe work builds on delayed streams modeling, the leading approach for streaming speech recognition, which forces a model to wait a fixed number of frames before committing to each word. The researchers argue that fixed wait time is a poor stand-in for how fast a transcript actually feels to a listener, because the same training transcript gets reused at every delay setting, needlessly holding back words the model already has enough audio to transcribe correctly. They introduce AWED, a metric that measures delay from the moment a word actually ends in the audio rather than from an arbitrary frame count, then retrain a streaming model with GRPO, a reinforcement learning method, using a reward that scores accuracy and that measured delay together. Tested against its own starting point and against Voxtral Realtime, a rival streaming system, the retrained model wins across every delay setting tried.\n\nAt the tightest lag tested, 80 milliseconds, word error rate dropped 30.8 percent relative to the baseline; at a more relaxed 480 milliseconds it still fell 5.7 percent, and median real-world delay dropped from 1.17 seconds to 1.04 seconds. That is a meaningful trade for captioning, voice assistants, and call center transcription, where the lag users actually notice is the one on the clock, not the frame count engineers optimize for.\n\nThe gains come from changing how the model is trained, not from bigger models or new chips. That is the more interesting story: sometimes the bottleneck was the reward function, not the architecture.","[\"speech-recognition\",\"streaming-asr\",\"ai-research\",\"reinforcement-learning\"]","2026-09-17T04:00:00.000Z","2026-09-18T02:40:40.392Z","2026-09-18T02:40:52.305Z","published",null,[],"ai",[26,27,28,29],"speech-recognition","streaming-asr","ai-research","reinforcement-learning",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.18333",0,{"sections":36},[37,41,45,50,55,59,63,68,73,77,82,87,92,97],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",3853,"2026-09-17T08:27:09.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":18},"Security","security",648,{"name":46,"slug":47,"count":48,"latest_published_at":49},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":18},"Hardware","hardware",154,{"name":60,"slug":61,"count":62,"latest_published_at":18},"Science","science",114,{"name":64,"slug":65,"count":66,"latest_published_at":67},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":69,"slug":70,"count":71,"latest_published_at":72},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":74,"slug":75,"count":76,"latest_published_at":18},"Dev Tools","dev-tools",73,{"name":78,"slug":79,"count":80,"latest_published_at":81},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":83,"slug":84,"count":85,"latest_published_at":86},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":88,"slug":89,"count":90,"latest_published_at":91},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":93,"slug":94,"count":95,"latest_published_at":96},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":98,"slug":99,"count":100,"latest_published_at":101},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]