Unifying Global and Near-Context Biasing in a Single Trie Pass

Unifying Global and Near-Context Biasing in a Single Trie Pass

Stať ve sborníku v databázi WoS či Scopus

Despite the success of end-to-end automatic speech recognition (ASR) models, challenges persist in recognizing rare, out-of-vocabulary wordsincluding named entities (NE)-and in adapting to new domains using only text data. This work presents a practical approach to address these challenges through an unexplored combination of an NE bias list and a word-level n-gram language model (LM). This solution balances simplicity and effectiveness, improving entities' recognition while maintaining or even enhancing overall ASR performance. We efficiently integrate this enriched biasing method into a transducer-based ASR system, enabling context adaptation with almost no computational overhead. We present our results on three datasets spanning four languages and compare them to state-of-the-art biasing strategies We demonstrate that the proposed combination of keyword biasing and n-gram LM improves entity recognition by up to 32% relative and reduces overall WER by up to a 12% relative.

Despite the success of end-to-end automatic speech recognition (ASR) models, challenges persist in recognizing rare, out-of-vocabulary wordsincluding named entities (NE)-and in adapting to new domains using only text data. This work presents a practical approach to address these challenges through an unexplored combination of an NE bias list and a word-level n-gram language model (LM). This solution balances simplicity and effectiveness, improving entities' recognition while maintaining or even enhancing overall ASR performance. We efficiently integrate this enriched biasing method into a transducer-based ASR system, enabling context adaptation with almost no computational overhead. We present our results on three datasets spanning four languages and compare them to state-of-the-art biasing strategies We demonstrate that the proposed combination of keyword biasing and n-gram LM improves entity recognition by up to 32% relative and reduces overall WER by up to a 12% relative.

Contextualisation and adaptation of ASR, real-time ASR, Aho-Corasick algorithm, Transformer-Transducer

Contextualisation and adaptation of ASR, real-time ASR, Aho-Corasick algorithm, Transformer-Transducer

THORBECKE, I.; VILLATORO-TELLO, E.; ZULUAGA, J.; KUMAR, S.; BURDISSO, S.; RANGAPPA, P.; CAROFILIS, A.; MADIKERI, S.; MOTLÍČEK, P.; PANDIA, K.; HACIOGLU, K.; STOLCKE, A.

2026

25.08.2026

Springer Nature

CHAM

978-3-032-02547-0

Lecture Notes in Artificial Intelligence

Lecture Notes in Computer Science

16029

Švýcarská konfederace

170

181

12

https://www.fit.vut.cz/research/group/speech/public/publi/2025/Iuliia_TSD2025_2025_co-author_Motlicek.pdf