--- title: How to Achieve 2.8x Faster Automatic Speech Recognition description: Standard RNN-Transducer (RNN-T) ASR decoders waste most of their compute stepping through silence one frame at a time. The Token-and-Duration Transducer (TDT)... canonical: https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How to Achieve 2.8x Faster Automatic Speech Recognition | daily.dev og:description: Standard RNN-Transducer (RNN-T) ASR decoders waste most of their compute stepping through silence one frame at a time. The Token-and-Duration Transducer (TDT)... og:url: https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet og:image: https://api.daily.dev/og/posts/ccu5h5JET.png og:image:alt: How to Achieve 2.8x Faster Automatic Speech Recognition og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How to Achieve 2.8x Faster Automatic Speech Recognition **[Daily Dose of Data Science \| Avi Chawla \| Substack](https://daily.dev/sources/dailydoseofds)** · 10 min read · 0 upvotes · 0 comments ## Summary Standard RNN-Transducer (RNN-T) ASR decoders waste most of their compute stepping through silence one frame at a time. The Token-and-Duration Transducer (TDT) fixes this by adding a second output head to the joint network that predicts how many frames to skip, not just which token to emit. This lets the decoder jump over silence in one step instead of confirming blank outputs repeatedly. The result is up to 2.82x faster decoding with equal or better word error rate, no encoder changes required. Speechmatics uses TDT in production, and NVIDIA's Parakeet TDT models top the HuggingFace Open ASR Leaderboard on throughput (RTFx) using the same encoder size as slower RNN-T competitors. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [How Together AI built the world’s fastest speech-to-text stack](https://daily.dev/posts/how-together-ai-built-the-world-s-fastest-speech-to-text-stack-zje9zzhdu) · Together AI · 0 upvotes · 0 comments - [Open ASR Leaderboard: Trends and Insights with New Multilingual & Long-Form Tracks](https://daily.dev/posts/open-asr-leaderboard-trends-and-insights-with-new-multilingual-long-form-tracks-gx0k7fwmm) · Hugging Face · 3 upvotes · 0 comments --- Tags: [#machine-learning](https://daily.dev/tags/machine-learning) [View this post on daily.dev](https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How to Achieve 2.8x Faster Automatic Speech Recognition","url":"https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet"},"datePublished":"2026-07-02T19:24:54.116Z","dateModified":"2026-07-02T19:26:49.110Z","description":"Standard RNN-Transducer (RNN-T) ASR decoders waste most of their compute stepping through silence one frame at a time. The Token-and-Duration Transducer (TDT)...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0145323bc5282672d9cceb7834fb7f99?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0145323bc5282672d9cceb7834fb7f99?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Daily Dose of Data Science | Avi Chawla | Substack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Daily Dose of Data Science | Avi Chawla | Substack","logo":"https://media.daily.dev/image/upload/s--4IHQgTOw--/f_auto/v1710503712/logos/dailydoseofds","url":"https://daily.dev/sources/dailydoseofds"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-to-achieve-2-8x-faster-automatic-speech-recognition-ccu5h5jet","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning","timeRequired":"PT10M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Daily Dose of Data Science | Avi Chawla | Substack","item":"https://daily.dev/sources/dailydoseofds"},{"@type":"ListItem","position":3,"name":"How to Achieve 2.8x Faster Automatic Speech Recognition"}]} ```