[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273872-105":59,"doc-detail-273872-en":130},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":123,"head_meta":125,"extra_data":127,"updated_unix":129},105,"en","turbobias-20-streaming-context-biasing-for-production-efficient-asr-systems","TurboBias 2.0 - Streaming Context-Biasing for Production-Efficient ASR Systems","","Contextualization is essential for production automatic speech recognition (ASR) systems, where user-provided phrases must be recognized accurately under strict latency constraints. While existing context-biasing techniques can raise accuracy, they often lack practical support for streaming inference, efficient batched decoding, per-user context lists, and low runtime overhead. TurboBias 2.0 delivers a production-oriented GPU framework for efficient phrase boosting in transducer-based ASR, enabling independent per-stream configurations without mixing contexts.",{"@graph":69,"@context":122},[70,84,105],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/turbobias-20-streaming-context-biasing-for-production-efficient-asr-systems/273872/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/turbobias-20-streaming-context-biasing-for-production-efficient-asr-systems/273872.png","ImageObject",300,407,{"name":92,"@type":93},"Ophelia","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":52},"InteractionCounter",{"@type":104},"ViewAction",{"@type":106,"mainEntity":107},"FAQPage",[108,114,118],{"name":109,"@type":110,"acceptedAnswer":111},"What problem does TurboBias 2.0 address in production ASR?","Question",{"text":112,"@type":113},"It targets the gap between accuracy-oriented context biasing methods and production needs such as streaming inference, batched decoding efficiency, per-user context lists, and low runtime overhead.","Answer",{"name":115,"@type":110,"acceptedAnswer":116},"How does TurboBias 2.0 enable personalized context biasing for multiple users?",{"text":117,"@type":113},"It uses per-stream batched decoding so each utterance in a batch can apply an independent context-biasing configuration, avoiding distractors and false accepts from shared trees.",{"name":119,"@type":110,"acceptedAnswer":120},"Does TurboBias 2.0 support both offline and streaming recognition?",{"text":121,"@type":113},"Yes. The framework supports both offline and streaming inference and can be used with greedy and beam-search decoding while keeping low latency and high throughput.","https://schema.org",{"og:url":83,"og:type":124,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":126,"canonical":83},"index,follow",{"doc_id":128,"site_id":62},273872,1789473095,{"code":4,"msg":5,"data":131},{"doc_id":128,"user_id":132,"nickname":92,"user_avatar":133,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":52,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":34,"language":139,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":140,"faqs":141,"seo_title":142,"seo_description":67,"update_tm":129,"read_time":143,"preview_page_count":4},7971461741311,"https://ap-avatar.wpscdn.com/avatar/74000253aff267980c6?x-image-process=image/resize,m_fixed,w_180,h_180&k=1779345379180704826","TurboBias 2.0: Streaming Context-Biasing for Production-Efficient ASR Systems  \nVladimir Bataev∗‡§ , Lilit Grigoryan∗‡, Andrei Andrusenko∗‡, Nikolay Karpov∗ , Vitaly Lavrukhin†, Boris Ginsburg†  \n∗NVIDIA, Yerevan, Armenia †NVIDIA, Santa Clara, USA  \n‡Equal contribution§ Corresponding author, [vbataev@nvidia.com](vbataev@nvidia.com)  \narXiv : 2608 .2 1343v 1 [ee ss .AS] 2 1 Aug 2026  \nAbstract—Contextualization is essential for production automatic speech recognition (ASR) systems, where user-provided phrases must be recognized accurately under strict latency constraints. Although many context-biasing methods improve recognition accuracy, they often do not address the practical requirements of modern production ASR systems: streaming inference, efficient batched decoding, user-specific context lists, and low runtime overhead. We propose TurboBias 2.0, a production-oriented framework for efficient phrase boosting in Transducer-based ASR systems. The framework extends GPUaccelerated TurboBias with a case-insensitive boosting graph and per-stream batched decoding, allowing each utterance in a batch to use an independent context-biasing configuration. This enables personalized context biasing for multiple simultaneous users without sharing or mixing their context lists. The proposed framework supports both offline and streaming inference and can be used with greedy and beam-search decoding. Experiments show that TurboBias 2.0 improves contextual phrase recognition while preserving low latency and high throughput.  \nIndex Terms—ASR, speech recognition, context-biasing, word boosting, phrase boosting, Transducers, beam search  \nI. INTRODUCTION  \nModern automatic speech recognition (ASR) systems achieve strong accuracy on general-purpose benchmarks, but they still often struggle with rare, domain-specific, and userdefined terminology. These terms may be infrequent in training data, yet they are often crucial for transcript usability. Therefore, context biasing remains an important component of production ASR systems, especially for meetings, earnings calls, medical transcription, and voice assistants.  \nA wide range of context-biasing approaches has been proposed. Deep-fusion methods inject contextual information directly into the ASR model through additional attention mechanisms, adapters, or specialized training objectives [1]–[5] . These methods can be accurate but often require model modifications and retraining. Recent speech and speechlanguage models can also use textual prompts with contextual phrases [6], [7], but their behavior may be sensitive to prompt design, context-list length, and latency constraints. In contrast, shallow-fusion and keyword-boosting methods modify decoding scores for hypotheses that match a predefined context list [8]–[11] . This makes them attractive for production because they can often be applied without retraining the ASR model.  \nSeveral recent works have focused on efficient contextual ASR. CTC-based Word Spotter (CTC-WS) detects keywords by matching CTC log-probabilities against a context graph and  \nthen correcting corresponding spans in the greedy ASR output [12] . While effective, this approach requires a CTC scoring path and a post-recognition correction stage. For Transducerbased ASR (RNNT [13], TDT [14]), it is most natural for hybrid Transducer-CTC models or pipelines with a separate CTC model. TurboBias introduced a GPU-accelerated phraseboosting tree for shallow-fusion context biasing and demonstrated efficient support for CTC, Transducer, and Attention Encoder-Decoder models [15] . More recently, Contextual Earnings-22 introduced a public contextual ASR benchmark with realistic custom vocabulary from earnings calls and evaluation settings with both local context and global context lists containing distractors [16] . Despite this progress, several requirements of high-load streaming production ASR remain insufficiently addressed.  \nFirst, modern ASR systems often produce formatted transc","cbCaiayC1ta4sMHM","https://ap.wps.com/l/cbCaiayC1ta4sMHM","pdf",662754,"English","# Introduction\n## Motivation for production-ready context biasing\n## Prior approaches and efficiency challenges\n## Proposed TurboBias 2.0 framework","[{\"question\":\"What problem does TurboBias 2.0 address in production ASR?\",\"answer\":\"It targets the gap between accuracy-oriented context biasing methods and production needs such as streaming inference, batched decoding efficiency, per-user context lists, and low runtime overhead.\"},{\"question\":\"How does TurboBias 2.0 enable personalized context biasing for multiple users?\",\"answer\":\"It uses per-stream batched decoding so each utterance in a batch can apply an independent context-biasing configuration, avoiding distractors and false accepts from shared trees.\"},{\"question\":\"Does TurboBias 2.0 support both offline and streaming recognition?\",\"answer\":\"Yes. The framework supports both offline and streaming inference and can be used with greedy and beam-search decoding while keeping low latency and high throughput.\"}]","TurboBias 2.0 - Streaming Context-Biasing for Production-Efficient ASR Systems | PDF",18]