Geographic Language Models for Automatic Speech Recognition

Improving Speech Recognition Accuracy of Local POI Using Geographical Models

“…An efficient way to improve speech recognition accuracy of POI names is to utilize geo-location dependent LMs [8,9,10,11]. For each user, Sten et al [9] trains a Geo-LM dynamically using nearby POI names and combines the Geo-LM with a baseline LM before or at decoding.…”

“…For each user, Sten et al [9] trains a Geo-LM dynamically using nearby POI names and combines the Geo-LM with a baseline LM before or at decoding. In [11], a class-based Geo-LM is constructed dynamically for each user depending on users' geographic location, within a difference-LM based weighted finite state transducer (WFST) system. All above approaches construct LMs or WFSTs on-the-fly according to users' geographical locations, which is time consuming and hard to incorporate plenty of POI names into a Geo-LM.…”

Predicting Entity Popularity to Improve Spoken Entity Recognition by Virtual Assistants

“…More specifically, knowledge about the user's interestswhich may eventually lead to an interaction with a specific intentcan be helpful to improve user experience. Xiao et al [39] improve ASR by bucketing users according to their coarse geographic location and enable region-specific query LMs during the ASR decoding process. By personalizing the ASR query model based on user location, they show a significant improvement in the accurate recognition of spoken point-of-interest queries.…”

“…3.1.1 Improving the ASR decoding process. At recognition time, contextual signals, such as partial recognition hypotheses [25] or the user location [39], can be used to modify the search space. Pusateri et al [25] combine multiple domain-specific expert n-gram LMs into a single LM by weighing the expert LMs based on the confidence expressed by each expert LM on how well they support specific left spoken contexts.…”