2020/07/23 by Kyung-Min Lee, Chiyoun Park, Lee, Kyungmin +7
Computer Science · #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Natural Language Processing Techniques #Speech Recognition and Synthesis #Topic Modeling
paper · pdf · doi:10.48550/arxiv.2007.11794
openalex publication_date 2020/07/23 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Recurrent Neural Network Language Models (RNNLMs) have started to be used in\nvarious fields of speech recognition due to their outstanding performance.\nHowever, the high computational complexity of RNNLMs has been a hurdle in\napplying the RNNLM to a real-time Large Vocabulary Continuous Speech\nRecognition (LVCSR). In order to accelerate the speed of RNNLM-based network\nsearches during decoding, we apply the General Purpose Graphic Processing Units\n(GPGPUs). This paper proposes a novel method of applying GPGPUs to RNNLM-based\ngraph traversals. We have achieved our goal by reducing redundant computations\non CPUs and amount of transfer between GPGPUs and CPUs. The proposed approach\nwas evaluated on both WSJ corpus and in-house data. Experiments shows that the\nproposed approach achieves the real-time speed in various circumstances while\nmaintaining the Word Error Rate (WER) to be relatively 10% lower than that of\nn-gram models.\n