基于小波基的光场数据 冗余性分析及重构算法的研究 - mycourses.pdfVIP

ROBUST AUTOMATIC SPEECH RECOGNITION R. Gao, J. Kallas, T. Langerak Dec. 9, 2016 CONTENT Introduction Related work Our methods 鈥 Mel-frequency cepstral coefficient(MFFC) 鈥 Perceptual Linear Prediction (PLP) 鈥 Spectral subtraction Results 2 Conclusion Discussion INTRODUCTION 飩n ASR system converts the speech signal into words/ linguistic entities by means of algorithms; 飩pplication Areas (demo): 飩 Voice search and interactions with mobile devices (e.g., Siri, Bing, Google) 飩 Voice control in home entertainment systems (e.g., Kinect on xBox) 飩 Noises can be simultaneous with speech; 飩 Non-speech artifacts can be extremely varied 飩 Background noises (music, dog bark, door slam, etc.) 飩 Microphone and channel noises (clicks, beeps, static, etc.) 飩 Non-lexical speaker noises (cough, laugh, lip smack, etc.) Clean data 3 Noisy data MOTIVATIONS AND TASKS 飩 Some reasons for errors 飩 Difficult acoustic conditions or background noises; 飩 Insufficient amount of training data; 飩 Acoustic mismatch between training and testing data. 飩he goal is to make automatic speech recognition method robust to a noisy environment. 飩asks 飩 1st Subtract static spectral components; 飩 2nd Build and compare different feature extraction methods; 飩 3rd Employ audio enhancing methods to improve results. 4 DATASE