Key Information Extraction from Broadcast in Traffic Domain

J. B. Gu,Y. F. Xie,Y. Y. Wang,X. Xie

DOI: https://doi.org/10.2991/aiie-15.2015.64

2015-01-01

Abstract:With the development of Artificial Intelligence, Automatic Speech Recognition has become an important field of cognitive computing in terms of intelligent vehicle. This paper introduces the application of Speech Recognition and Natural Language Processing on intelligent vehicle that extracts key information from traffic broadcast. This system consists of three modules, speech detection module, keywords output module, and structured information extraction module. The speech detection module uses Non-negative matrix factorization to make speech enhancement, which separates the speech from the background music. Using Mel-frequency Cepstral Coefficients (MFCCs) extracted from the speech as features, it trains the model using Hidden Markov Model (HMM). Language model like n-gram improves the performance of the recognizer. In structured information extraction module, Conditional random field (CRF) is used to make partial parsing. Then it fills the slots, and outputs structured traffic condition information. We give the experiment results of every module.