Noise robust speech recognition using deep neural network

... possible noise at the near end of the speech 13 NOISE- ROBUST SPEECH RECOGNITION Ambient Noise zenv Lombard Effect Speaker Stress/ Workload x Clean Speech Additive Transmission Noise ztrans Reciever Noise ... DNN-based Speech Recognition, submitted to Interspeech, ISCA, 2014 • Bo Li, Khe Chai Sim; An Ideal Hidden-Activation Mask for Deep Neural Networks based Noise-...

Ngày tải lên: 09/09/2015, 11:23

... directional noise sources However, it tends to be less effective for omnidirectional noises In order to make the speech recognition more robust in a variety of noise environments, we added hidden Markov ... microphone array was able to achieve almost the same recognition accuracies as those of the headset microphone Conclusions We developed a noise robust speech recogn...

Ngày tải lên: 21/06/2014, 20:20

... constructed clean -speech HMMs and an HMM for sudden noise The recognition units in clean -speech HMMs were triphones, which were trained using cleanspeech data An HMM for sudden noise was trained using sudden ... 5(1a) We compared the recognition accuracies of clean -speech HMMs without Δ features, clean -speech HMMs with Δ features, FHMMs with Δ features, and FHMMs w...

Ngày tải lên: 22/06/2014, 23:20

... be compared to spectral subtraction Evaluation database As test material we took the resource management (RM) database (available from LDC [34]) These data are considered as clean data, to which ... singular value of Σ The enhanced signal s(k) is recovered by averaging along the antidiagonals of Hs Dologlou and Carayannis [17], and later on Hansen and Jensen [18] proved that t...

Ngày tải lên: 22/06/2014, 23:20

... scale up to large speech recognition tasks This thesis demonstrates that neural networks can indeed form the basis for a general purpose speech recognition system, and that neural networks offer ... and a summary of related work in speech recognition and neural networks: • Chapter reviews the field of speech recognition • Chapter reviews the field of neural network...

Ngày tải lên: 28/04/2014, 10:18

... normalization for noise robust speech recognition,” Speech Communication, vol 25, no 1–3, pp 133–147, 1998 [5] C Kermorvant, “A comparison of noise reduction techniques for robust speech recognition,” ... Ney, “Quantile based histogram equalization for noise robust speech recognition,” in Proceedings of the 7th European Conference on Speech Communication and Technolo...

Ngày tải lên: 22/06/2014, 19:20

... This article presents a novel noise robust speech parameterization procedure WPDAM based on wavelet packet decomposition ASR performance evaluation using the Aurora database shows the efficiency and ... WPDAM and AFE on the Aurora database can be seen from Tables and Tables 5, 6, and present the comparison between WPDAM and AFE on the Aurora database When compared to AFE, the...

Ngày tải lên: 22/06/2014, 20:20

... .1 1 .1 Speech Recognition 1. 2 Neural Networks .4 1. 3 Thesis Outline ... of speech recognition • Chapter reviews the field of neural networks • Chapter reviews the intersection of these two fields, summarizing both past and present approaches to speech recognition using ... and discontinuous speech recognition is relatively easy because word boundaries ar...

Ngày tải lên: 13/08/2014, 02:21

... 12 word (unlimited) MARKET triphone (10000) MA,MAR,ARK,RKE,KET,ET$ $ senone (4000) M = 3843 ,22 57,1056; A = 1894, 124 7,38 52; generalized triphone (4000) 1087,486 ,25 02, 986,3814 ,27 15 diphone (20 00) ... model 26 Review of Speech Recognition 2. 3.4 Limitations of HMMs Despite their state-of-the-art performance, HMMs are handicapped by several well-known weaknesses, namely: • The F...

Ngày tải lên: 13/08/2014, 02:21

... In this case, the net input is given by: 3. 2 Fundamentals of Neural Networks 31 y1 y1 y2 * wj1 y2 wj2 wj1 xj yj xj yj wj3 y3 y3 (a) * wj2 (b) y4 Figure 3. 2: Computing unit activations: x=net input, ... occur in time Assuming the task is speech recognition, or some other task in the temporal domain 3. 3 A Taxonomy of Neural Networks 39 The TDNN is trained using standard bac...

Ngày tải lên: 13/08/2014, 02:21

... infancy, and it is premature to rely on neural networks for temporal modeling in a speech recognition system 4. 3 NN-HMM Hybrids We have seen that neural networks are excellent at acoustic modeling ... which the speech frames are produced by a combination of signal analysis 4. 3 NN-HMM Hybrids 63 and neural networks; the speech frames then serve as inputs for an ordinary...

Ngày tải lên: 13/08/2014, 02:21

... speaks no English Janus performs speech translation by integrating three modules — speech recognition, text translation, and speech generation — into a single end-to-end system Each of these modules ... three dialogs (41 sentences) using a reduced vocabulary without a grammar The Conference Registration database was developed in conjunction with the Janus Speech- to -Speech Transla...

Ngày tải lên: 13/08/2014, 02:21

... this chapter perplexity System HMM-1 HMM-5 HMM-10 LVQ LPNN 111 402 96% 97% 98% 97% 55% 70% 75% 80% 60 % 58% 66 % 74% 40% Table 6. 4: Word accuracy of HMM-n with n mixture densities, LVQ, and LPNN 6. 4 ... Continuous speech (P=7) Continuous speech (P=402) 80 960 42080 64 66 70% 91% 14% 67 % 91% 20% 39% n/a n/a Table 6. 6: Results of Hidden Control experiments Parameter sharing mu...

Ngày tải lên: 13/08/2014, 02:21

... Figure 7. 10 7 Classification Networks 114 1 symmetric sigmoid sigmoid y = – 1 + e –x y = -1 + e –x -1 ∑ yi -1 softmax = i x e i y i = -xj ∑e -1 j -1 y = ( x ) = - – ... inputs 71 .8% 74 .9% 80.4% 74 .5% 67. 2% 82.0% 72 .9% 86.9% 88.2% 79 .0% 89.2% 86.6% 89.3% 86.0% 90.9% 91.5% 89.6% 86.0% 77 .9% 88.0% 81.4% 90.6% 90 .7% 82.0% 90.5% 88.2% 90.4% 86.6% 68% 66% 47% 4...

Ngày tải lên: 13/08/2014, 02:21

... NN-HMM 41,000 61 Feb89+Oct89 89 .2% MS-TDNN NN-HMM 67,000 61 Feb89+Oct89 90.5% MLP (ICSI) NN-HMM 156,000 69 Feb89+Oct89 87 .2% CI-Sphinx HMM 111,000 48 Mar 88 84.4% CI-Decipher HMM 126,000 69 Feb89+Oct89 ... 8 Comparisons 1 48 perplexity System HMM-1 111 402(a) test on training set 402(b) 111 55% HMM-5 96% 71% 58% 76% HMM-10 97% 75% 66% 82 % LPNN 97% 60% 41% HCNN 75% LVQ 98% 84 % 74%...

Ngày tải lên: 13/08/2014, 02:21