Page 95 - 201903

第 38 卷 第 3 期 李云红等： 一种改进的 DNN-HMM 的语音识别方法 377

[8] Wang J, Li L, Wang D, et al. Research on generalization gressive moving average spectrogram features and chan-
property of time-varying Fbank-weighted MFCC for I- nel dropout[J]. Pattern Recognition Letters, 2017, 100(1):
vector based speaker verification[C]// International Sym- 44–50.
posium on Chinese Spoken Language Processing. IEEE, [13] Miloslavskaya V, Trifonov P. Sequential decoding of po-
2014: 423–423. lar codes[J]. IEEE Communications Letters, 2014, 18(7):
[9] Hinton G E, Osindero S, Teh Y W. A fast learning algo- 1127–1130.
rithm for deep belief nets[J]. Neural Computation, 2006, [14] Jiang Y B, Gao X N, Han P. Application of maximum en-
18(7): 1527–1554. tropy model in distribution of vehicle speed[J]. Advanced
[10] Mohamed A R, Dahl G, Hinton G. Acoustic modeling us- Materials Research, 2015, 1079–1080: 942–945.
ing deep belief networks[J]. IEEE Transactions on Audio, [15] Lopes C, Perdigao F. Phone Recognition on the TIMIT
Speech and Language Processing, 2012, 20(1): 14–22. Dataset[M]//Speech Technologies. InTech, 2011.
[11] 张劲松, 高迎明, 解焱陆. 基于 DNN 的发音偏误趋势检测 [J]. [16] Salakhutdinov R, Hinton G E. Deep Boltzmann ma-
清华大学学报: 自然科学版, 2016, 56(11): 1220–1225. chines[C]//Proceedings of International Conference on
Zhang Jinsong, Gao Yingming, Xie Yanlu. Mispronunci- Artificial Intelligence and Statistics 2009. Brookline, MA,
ation tendency detection using deep neural networks[J]. USA: Microtome Publishing, 2009: 448–455.
Journal of Tsinghua University(Science and Technology), [17] Hu W J, Fu M J, Pan W L. Primi speech recognition
2016, 56(11): 1220–1225. based on deep neural network[C]//Proceedings of IEEE
[12] Kovacs G, Toth L, Compernolle D V, et al. Increas- International Conference on Intelligent Systems. Wash-
ing the robustness of CNN acoustic models using autore- ington D.C., USA: IEEE Press, 2016: 667–671.