[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-researchers-split-compression-by-layer-to-speed-up-speech-ai":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},6430,"researchers-split-compression-by-layer-to-speed-up-speech-ai","Researchers Split Compression by Layer to Speed Up Speech AI","A new codec called LACE compresses audio data differently at each processing layer, a tweak that speeds up text-to-speech systems without hurting sound quality.","A codec used in speech AI research just got a targeted efficiency upgrade, and it comes from rethinking a step most systems treat as one-size-fits-all.\n\nResearchers have proposed LACE (Layer-Adaptive Codec Encoding), a neural audio codec that compresses each quantization layer separately instead of forcing every layer to share the same compression boundaries. That distinction matters because these codecs typically use multiple stacked layers to represent audio, and each layer changes at a different pace over time. Prior \"dynamic frame rate\" codecs applied one compression pass across all layers, which meant faster-changing layers and slower-changing ones got squeezed on the same schedule. LACE also adds two mechanisms, union alignment and boundary anchors, so the layers stay in sync for text-to-speech use even after being compressed independently. Tests on the LibriTTS dataset showed LACE beats earlier dynamic frame rate methods on the tradeoff between compression rate and audio quality, and it makes text-to-speech inference more efficient without a meaningful quality hit.\n\nThe real audience here isn't audio engineers chasing marginal gains, it's anyone building voice AI at scale. Speech codecs feed directly into speech language models, and frame rate is basically the tax those models pay per second of audio processed. A method that shaves that tax while holding quality steady lowers compute costs for every downstream voice application built on top of it.\n\nThis is an incremental fix, not a new paradigm, but incremental fixes are how the audio-codec pipeline behind voice assistants and TTS products quietly gets cheaper to run. The code has been released as part of the ESPnet3 codec recipe, so the harder question is now whether it survives contact with production-scale voice systems rather than a single benchmark dataset.","[\"speech ai\",\"audio codecs\",\"text-to-speech\",\"research\"]","2026-09-16T04:00:00.000Z","2026-09-17T15:58:09.120Z","2026-09-17T15:58:21.032Z","published",null,[],"ai",[26,27,28,29],"speech ai","audio codecs","text-to-speech","research",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.17509",0,{"sections":36},[37,41,46,51,56,60,64,69,74,78,83,88,93,98],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",3852,"2026-09-17T08:27:09.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",648,"2026-09-17T04:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":45},"Hardware","hardware",154,{"name":61,"slug":62,"count":63,"latest_published_at":45},"Science","science",114,{"name":65,"slug":66,"count":67,"latest_published_at":68},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":45},"Dev Tools","dev-tools",73,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]