基于时域建模的自动语音识别-计算机工程与应用.PDF

Computer Engineering and Applications 计算机工程与应用 2017 ，53（20 ） 243 基于时域建模的自动语音识别 用 王海坤，伍大勇，刘 江，王士进，胡国平，胡 郁 应 WANG Haikun, WU Dayong, LIU Jiang, WANG Shijin, HU Guoping, HU Yu 与 科大讯飞股份有限公司 研究院，合肥 230088 程 Research of IFLYTEK CO., LTD, Hefei 230088, China g 工 r o WANG Haikun, WU Dayong, LIU Jiang, et al. Automatic speech recognition based on time domain modeling. Com- . j puter Engineering and Applications, 2017, 53 （20 ）：243-248. 机 a 算 e Abstract ：End-to-end neural networks can automatically learn feature transformation from original data, which can solve c . the mismatch between hand designed features and specific tasks. The traditional end-to-end neural network for speech rec- 计 w ognition uses a time domain convolution network as the feature extraction model, recurrent neural network and full con- nected feed-forwarddeep neural network as the acoustic model, which has some limitations in performance and efficiency. w From the aspects of the performanceof thefeature extraction module and the training efficiency of the acoustic model, an w end-to-end speech recognition model combining the multi-time and frequency resolution convolution and the feed- forward neural network with memory modules is proposed. On the real recording test dataset, the proposed method reduces the word error rate by 10%, training time by 80% compared with the traditional method. Key words ：convolution neural network; recurrent neural network; acoustic model; end-to-end neural network 摘 要：端到端神经