2020/08/18 by Rishika Agarwal, Xiaochuan Niu, Agarwal, Rishika +9
Computer Science · #Adversarial Robustness in Machine Learning #Audio and Speech Processing (eess.AS) #Computation and Language (cs.CL) #FOS: Computer and information sciences #FOS: Electrical engineering #Hate Speech and Cyberbullying Detection #Internet Traffic Analysis and Secure E-voting #Machine Learning (cs.LG) #Sound (cs.SD) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2008.08113
openalex publication_date 2020/08/18 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
False triggers in voice assistants are unintended invocations of the\nassistant, which not only degrade the user experience but may also compromise\nprivacy. False trigger mitigation (FTM) is a process to detect the false\ntrigger events and respond appropriately to the user. In this paper, we propose\na novel solution to the FTM problem by introducing a parallel ASR decoding\nprocess with a special language model trained from "out-of-domain" data\nsources. Such language model is complementary to the existing language model\noptimized for the assistant task. A bidirectional lattice RNN (Bi-LRNN)\nclassifier trained from the lattices generated by the complementary language\nmodel shows a 38.34 % relative reduction of the false trigger (FT) rate at\nthe fixed rate of 0.4 % false suppression (FS) of correct invocations,\ncompared to the current Bi-LRNN model. In addition, we propose to train a\nparallel Bi-LRNN model based on the decoding lattices from both language\nmodels, and examine various ways of implementation. The resulting model leads\nto further reduction in the false trigger rate by 10.8 %.\n