Search
Search Funnelback University
- Refined by:
- Date: 2015
1 -
30 of
30
search results for news |u:mi.eng.cam.ac.uk
Fully-matching results
-
mi.eng.cam.ac.uk/research/dialogue/EPSRCProj/assets/xml/rss.xsl
mi.eng.cam.ac.uk/research/dialogue/EPSRCProj/assets/xml/rss.xsl18 Sep 2015: It will be updated periodically in your reader. New to feeds? -
lect14
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect14.2up.pdf10 Nov 2015: speaker tracking in broadcast news transcription: text in-dependent, speaker identification. ... expectation maximisation (EM) can be used to train themodel. For the “new” model mean. -
lect3.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect3.pdf10 Nov 2015: We now have new estimates of the component parameters. We can therefore get new estimates of the assignment z. -
lect12
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect12.2up.pdf10 Nov 2015: no data in the right node. 6. recursively apply the above algorithm to each new node. ... Heredecision do not have to be parallel to the axis. The questionis how to select the direction of the new split. -
lect13
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect13.2up.pdf10 Nov 2015: 11. Non-Parameteric Techniques 3. Non-Parametric Density Estimation. The histogram can be generalised to other forms of region.The probability that a new vector x drawn from the unknowndensity -
lect1.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect1.pdf10 Nov 2015: Sponsored Link. Sign in. Web Images News Maps Products Groups. Search: the web pages from the UK New! ... viavoice imageGoogle. New! • Search engines are essential. – query to find previous image viavoice image. – -
lect3_pres.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect3_pres.pdf10 Nov 2015: ĉ1 =n1. n1 n2. Similarly for class ω2. We now have new estimates of the component parameters.We can therefore get new estimates of the assignment z. -
lect12.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect12_pres.pdf25 Nov 2015: no data in the right node. 6. recursively apply the above algorithm to each new node. ... Heredecision do not have to be parallel to the axis. The questionis how to select the direction of the new split. -
lect1.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect1_pres.pdf10 Nov 2015: Sponsored Link. Sign in. Web Images News Maps Products Groups. Search: the web pages from the UK New! ... viavoice imageGoogle. New! 3. 4F10: Statistical Pattern Processing. Rafales de marque - lecteur dans la technologie de. -
lect13.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect13_pres.pdf25 Nov 2015: Non-Parameteric Techniques 3. Non-Parametric Density Estimation. The histogram can be generalised to other forms of region.The probability that a new vector x drawn from the unknowndensity p(x) -
A LANGUAGE SPACE REPRESENTATION FOR SPEECH RECOGNITION
mi.eng.cam.ac.uk/~mjfg/icassp15-ragni.pdf18 May 2015: Such an approach could not only re-duce development costs for each new language but also provide au-tomatic means for language analysis. ... Each language l =1,. ,L defines a new cluster. Trees for non-bias clusters are ini-tialised to have zero mean -
lect4.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect4.pdf10 Nov 2015: We would. like a new estimate so that for the parameters at the k 1th. ... log-likelihood of the complete dataset in terms of the new. model parameters, θ(k1),. -
Improving Speech Recognition and Keyword Searchfor Low Resource…
mi.eng.cam.ac.uk/~ar527/mendels_is2015a.pdf3 Jun 2015: Wescrape text from multiple genres including blogs, online news,translated TED talks, and subtitles. ... source. Some of the sources were multilingual, suchas TED talks, and others were single language, such as blogsand news sources. -
A LANGUAGE SPACE REPRESENTATION FOR SPEECH RECOGNITION A. Ragni, ...
mi.eng.cam.ac.uk/~ar527/ragni_icassp2015a.pdf6 Apr 2015: Such an approach could not only re-duce development costs for each new language but also provide au-tomatic means for language analysis. ... Each language l =1,. ,L defines a new cluster. Trees for non-bias clusters are ini-tialised to have zero mean -
Improving Speech Recognition and Keyword Searchfor Low Resource…
mi.eng.cam.ac.uk/~mjfg/interspeech15-mendels-babel.pdf9 Oct 2015: Wescrape text from multiple genres including blogs, online news,translated TED talks, and subtitles. ... source. Some of the sources were multilingual, suchas TED talks, and others were single language, such as blogsand news sources. -
Joint Decoding of Tandem and Hybrid Systems for Improved ...
mi.eng.cam.ac.uk/~ar527/wang_is2015a.pdf3 Jun 2015: The acoustic log-likelihoods fromthe tandem GMM acoustic model and hybrid DNN are linearlycombined as new acoustic scores, which are used in the Viterbidecoding process. ... Tyagi, “New entropy based combi-nation rules in HMM/ANN multi-stream ASR,” -
lect4.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect4_pres.pdf10 Nov 2015: We wouldlike a new estimate so that for the parameters at the k 1th. ... dzi. The second term is not dependent on the new model param-eters, the distribution of zi is known. -
lect8.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect8_pres.pdf20 Nov 2015: Support Vector Machines For Texture Classification. • ewsRec, a SVM-driven Personal Recommendation Sys-tem for News Websites. • -
Association Between Femur Size and a FocalDefect of the ...
mi.eng.cam.ac.uk/reports/svr-ftp/gee_tr696.pdf3 Feb 2015: InSection 4, we discuss some important details of the methodology and also how the results shed new light onthe mechanisms by which femur size affects fracture risk. -
Joint Decoding of Tandem and Hybrid Systems for Improved ...
mi.eng.cam.ac.uk/~mjfg/interspeech15-wang-babel.pdf9 Oct 2015: The acousticlog-likelihoods from the tandem GMM acoustic model and hy-brid DNN are linearly combined as new acoustic scores, whichare used in the Viterbi decoding process. ... Tyagi, “New entropy based combi-nation rules in HMM/ANN multi-stream ASR,” -
lect9_pres
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect9_pres.2up.pdf10 Nov 2015: Support Vector Machines For Texture Classification. • ewsRec, a SVM-driven Personal Recommendation Sys-tem for News Websites. • -
Annotating large lattices with the exact word error Rogier ...
mi.eng.cam.ac.uk/~mjfg/Kernel/van_dalen-2015-exact_error.pdf16 Jun 2015: This approximation becomes particularly problematicwith new types of acoustic models that require flexible align-ments. ... 4. ResultsTo test the new algorithm for error-marking lattices, a realistictraining set with long utterances was chosen. -
IMPROVING MULTIPLE-CROWD-SOURCED TRANSCRIPTIONS USING A…
mi.eng.cam.ac.uk/~mjfg/ALTA/publications/van_dalen-2015-improving.pdf11 Feb 2015: and a new system is built, applying minimum phone errortraining and speaker-adaptive training. ... 14] Daniel Jurafsky and James H. Martin, Speech and LanguageProcessing, Prentice-Hall, New Jersey, 2000. -
Department of Engineering 1 Monoids: e�cient segmental features…
mi.eng.cam.ac.uk/~mjfg/Kernel/van_dalen-2013-tr-monoid_features17 Apr 2015: Amore realistic strategy for decoding would prune the hypothesis space. is reporttherefore proposes a new, more exible class of features. ... e new features are as ecient as features used in Ragniand Gales (2012); van Dalen et al. -
main
mi.eng.cam.ac.uk/~cipolla/publications/article/2016-CVIU-EVTTS.pdf10 Sep 2015: New interactive technologies provide an opportunity for ASC individuals61to practice their communication skills. ... Errorsare decreased with each new iteration of the model. (a) (b) (c) (d) (e) (f) (g). -
main
mi.eng.cam.ac.uk/~cipolla/archive/Publications/article/2016-CVIU-EVTTS.pdf10 Sep 2015: New interactive technologies provide an opportunity for ASC individuals61to practice their communication skills. ... Errorsare decreased with each new iteration of the model. (a) (b) (c) (d) (e) (f) (g). -
Department of Engineering 1 Generative Kernels and Score-Spaces…
mi.eng.cam.ac.uk/~mjfg/Kernel/rcv25_2015_y3.pdf1 Jun 2015: e new features are as ecient as features used in Ragniand Gales (2012); van Dalen et al. ... Also, as new models for speechrecognition are becoming more complex and features richer, the xed time alignmentscould restrict performance. -
Model-based Approaches to Robust SpeechRecognition in Diverse…
mi.eng.cam.ac.uk/~mjfg/thesis_yw293.pdf27 Oct 2015: Two new. approaches to enforcing the independence constraint are proposed in Chapter 6. ... detect the beginning and the end of speech events; in a broadcast news transcription system,. -
"Refinements in Hierarchical Phrase-Based Translation…
mi.eng.cam.ac.uk/~wjb31/ppubs/jpino2015HieroRefinementsThesis.pdf6 Feb 2015: a new cabinet,responsible for handling daily affairs of the new government.”. ... The president accepted it and at the same time alsoasked him to form a new cabinet government to be responsiblefor routine administration.”. -
PROBABILISTIC ACOUSTIC MODELLING FOR PARAMETRIC SPEECH SYNTHESIS Sean …
mi.eng.cam.ac.uk/~wjb31/ppubs/shannon2014probabilistic-thesis.pdf2 Feb 2015: As we develop the new model and generation method we also investigate and develop. ... method. To address the second issue, we introduce a new GV-like generation method.
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.