Search
Search Funnelback University
- Refined by:
- Date: 2017
1 -
20 of
24
search results for TALK:Vb20 24 / |u:mi.eng.cam.ac.uk
where 0
match all words and 24
match some words.
Results that match 1 of 2 words
-
UKspeech2017
mi.eng.cam.ac.uk/UKSpeech2017/posters/y_wang.pdf17 Nov 2017: L1, proficiency level, recordingSpontaneous responses increase difficulty, e.g. disfluenciesTranscribing is challenging inter-annotator error rate about 24.7%. -
Deep Learning for Speech Recognition
mi.eng.cam.ac.uk/~mjfg/LxMLS17.pdf29 Nov 2017: Network Interpretation [24]. Standard /ay/ Stimulated /ay/. • Deep learning usually highly distributed - hard to interpret• awkward to adapt/understand/regularise• modify training - add stimulation regularisation• improves ASR performance. -
Template.dvi
mi.eng.cam.ac.uk/~ar527/chen_icassp2017a.pdf22 Mar 2017: Mongolian FLP 511K 24.0K - 4.19 12.19WEB 139M 199.8K 0.93 2.10 5.62. ... 24,no. 11, pp. 2146–2157, 2016. [15] Xie Chen, Yongqiang Wang, Xunying Liu, Mark Gales, andP. -
.poster_jeremy_v2.tex.dvi
mi.eng.cam.ac.uk/UKSpeech2017/posters/j_wong.pdf17 Nov 2017: 207Vseparate 45.8 46.0 46.6MT 47.7 47.8 47.3MT-TS 45.7 45.7 46.3. AMIseparate 24.5 24.6 24.6MT 25.4 25.5 ... 25.1MT-TS 24.3 24.4 24.6. -
STIMULATED TRAINING FOR AUTOMATIC SPEECH RECOGNITION ANDKEYWORD…
mi.eng.cam.ac.uk/~ar527/ragni_icassp2017b.pdf22 Mar 2017: These weretrained on FLP data of 24 Babel languages and CTS data of 4 addi-tional languages, English, Spanish, Arabic and Mandarin, releasedby LDC. ... Stacked Hybrids were trained withand without stimulated training using monophone initialisation -
Automa(c Analysis of Mo(va(onal Interviewing with Diabetes Pa(ents…
mi.eng.cam.ac.uk/UKSpeech2017/posters/x_wei.pdf20 Nov 2017: Results:. Model Senone No. WER (%). Baseline DNN-‐HMM 3981 53.13. MI adapted DNN 3981 47.24. ... Hhit NF NREF PRC RCL. lium 35 144 93 0.24 0.38 ivector 42 159 93 0.26 0.45. -
MORPH-TO-WORD TRANSDUCTION FOR ACCURATE AND EFFICIENT AUTOMATICSPEECH …
mi.eng.cam.ac.uk/~ar527/ragni_icassp2017a.pdf22 Mar 2017: FLP Web FLP Web (#) ASR KWSSwahili 294 – 24.4 0 8.2 8.5 19.6Dholuo 467 1,217 17.5 18.8 6.1 3.0 10.0Amharic 388 ... 4, pp. 1738–1752, 1990. [24] P. Ghahremani, B. BabaAli, D. Povey, K. -
Future Word Contexts in Neural Network Language Models
mi.eng.cam.ac.uk/UKSpeech2017/posters/x_chen.pdf17 Nov 2017: dev evalwords (w/s) PPL. ng4 - - 80.4 23.8 24.2uni-rnn - 4.5K 66.8 21.7 22.1. ... ng4 - 23.8 23.5 24.2 23.9uni-rnn - 21.7 21.5 21.9 21.7. -
Experimental Studies on Teacher-student Training of Deep Neural…
mi.eng.cam.ac.uk/UKSpeech2017/posters/q_li.pdf20 Nov 2017: 22. 23. 24. 25. 26. 27. 28. 29. 30. PER. (%). 3-layer (100) Baseline3-layer (100) Student3-layer (250) Baseline3-layer (250) Student3-layer (500) Baseline3-layer (500) Student4-layer (500) ... PER (%)7-layer (500) 24.55 23.55RNN 23.84 20.59Ensemble 23.73 -
Use of Graphemic Lexicons for Spoken Language Assessment
mi.eng.cam.ac.uk/UKSpeech2017/posters/k_knill.pdf17 Nov 2017: Decoder Gujarati Mixed(word) %PER %GER %PER %GER. Ph 25.8 24.9 33.9 32.9Gr 29.0 23.7 36.6 30.8. -
Modular Construction of Complex Deep Learning Architectures in HTK
mi.eng.cam.ac.uk/UKSpeech2017/posters/f_kreyssig.pdf20 Nov 2017: I All models used 24 log-Mel filter bank coefficients with their and values as input features, except the CNN which used40 without any. -
A learned emotion space for emotion recognition and emotive speech…
mi.eng.cam.ac.uk/UKSpeech2017/posters/z_hodari_poster.pdf23 Dec 2017: Table 1: Performance classifying; happy, sad, angry, neutral. Model Inputs AccuracyRandom N/A 24.14%Most common N/A 33.00%LSTM eGeMAPS LLDs 43.17%TD-CNN Spectrogram -
Template.dvi
mi.eng.cam.ac.uk/~mjfg/CUED-Chen-RNNLMKWS.pdf22 Mar 2017: Mongolian FLP 511K 24.0K - 4.19 12.19WEB 139M 199.8K 0.93 2.10 5.62. ... 24,no. 11, pp. 2146–2157, 2016. [15] Xie Chen, Yongqiang Wang, Xunying Liu, Mark Gales, andP. -
Low-Resource Speech Recognition and Keyword-Spotting
mi.eng.cam.ac.uk/~mjfg/SPECOM_2017.pdf29 Nov 2017: 23/63. Stimulated Systems. /ey//em/. /sil/. /sh/. /ow/ /ay/. 24/63. Stimulated Network Training. • -
MORPH-TO-WORD TRANSDUCTION FOR ACCURATE AND EFFICIENT AUTOMATICSPEECH …
mi.eng.cam.ac.uk/~mjfg/CUED-Ragni-Morph-To-Word.pdf22 Mar 2017: FLP Web FLP Web (#) ASR KWSSwahili 294 – 24.4 0 8.2 8.5 19.6Dholuo 467 1,217 17.5 18.8 6.1 3.0 10.0Amharic 388 ... 4, pp. 1738–1752, 1990. [24] P. Ghahremani, B. BabaAli, D. Povey, K. -
STIMULATED TRAINING FOR AUTOMATIC SPEECH RECOGNITION ANDKEYWORD…
mi.eng.cam.ac.uk/~mjfg/CUED-Ragni-Stimulated-ASR-KWS.pdf22 Mar 2017: These weretrained on FLP data of 24 Babel languages and CTS data of 4 addi-tional languages, English, Spanish, Arabic and Mandarin, releasedby LDC. ... Stacked Hybrids were trained withand without stimulated training using monophone initialisation -
1 Statistical Sequence ModellingMark Gales Speech recognition and…
mi.eng.cam.ac.uk/~mjfg/sequence17-draft.pdf24 May 2017: Tt=1. p(yt|ht,h̃t) (1.24). where the normalisation term ensures that this is valid PDF. ... 24. Fcml(λ;D) =n. i=1. log( p(w(i)1:L(i)|Y(i). 1:T (i) ; λ)) (1.100). -
How Does the Femoral Cortex Depend onBone Shape? A ...
mi.eng.cam.ac.uk/reports/svr-ftp/gee_tr704.pdf15 Jun 2017: How Does the Femoral Cortex Depend onBone Shape? A Methodology for the Joint. Analysis of Surface Texture and Shape. A. H. Gee, G. M. Treece and K. E. S. Poole. CUED/F-INFENG/TR 70415 June 2017. Cambridge University Engineering DepartmentTrumpington -
University of CambridgeEngineering Part IB Information Engineering…
mi.eng.cam.ac.uk/~cipolla/lectures/PartIB/old/2017-DNN-lecture-3.pdf18 May 2017: the 9 filters create 9 images, which have 9 24 24 = 5184pixels, and thus we need 51,840 parameters to reduce those. -
IB-interestpoints.dvi
mi.eng.cam.ac.uk/~cipolla/lectures/PartIB/old/2017-IB-handout2.pdf18 May 2017: outliers in the output of the corner detector. 24 Engineering Part IB: Paper 8 Image Matching.
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.