Search
Search Funnelback University
201 -
250 of
1,000
search results for katalk:za33 24 |u:mi.eng.cam.ac.uk
where 0
match all words and 1,000
match some words.
Results that match 1 of 2 words
-
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). -
Neural Belief Tracker: Data-Driven Dialogue State Tracking Nikola…
mi.eng.cam.ac.uk/~sjy/papers/mowt17.pdf20 Feb 2018: Computer Speech and Lan-guage 24:150–174. Xiaodong Zhang and Houfeng Wang. 2016. -
ENGINEERING TRIPOS PART IIB ELECTRICAL AND INFORMATION SCIENCES…
mi.eng.cam.ac.uk/~cipolla/resource/4F12exam.pdf12 Apr 2022: with (X, Y ) vertices (23, 3), (5, 30), (45, 24) and (15, 48), calcu-late bounds on s. -
THE DEVELOPMENT OF THE CAMBRIDGE UNIVERSITY RT-04 DIARISATION SYSTEM…
mi.eng.cam.ac.uk/reports/svr-ftp/tranter_rt04.pdf10 Jan 2005: show. (b) Mean DER across all 24 development shows whenonly including c0 in clustering if the mean value is above a criticalthreshold. ... RT-04 RT-04 RT-04 23.86 24.12. Table 12. Progress since RT-03s on theeval04f data. -
Progress in English Conversational TelephoneSpeech Transcription Khe…
mi.eng.cam.ac.uk/research/projects/EARS/pubs/sim_sttmar05.pdf12 Apr 2005: S1 6K (28)0. 34.1 26.0 30.2 26.4S4 9K (36). (ML)33.0 24.8 29.0 25.3. ... 32.3 22.9 27.8pMPEMPE 35.1 26.0 30.7 33.2 24.1 28.8 32.9 23.6 28.4. -
book.dvi
mi.eng.cam.ac.uk/~sjy/papers/wipy06.pdf20 Feb 2018: vn(s) = r(s, π̂(n)) γ. s′S. oO. p(s′|s, π̂(n))p(o|s′, π̂(n))vl(n,o)(s′) (8.24). ... This style of execution is distinct frompolicy iteration, in which the nodesand links of the controller are changed and the controller isre-evaluated -
Bin Jia, Khe Chai Sim et al: CU-HTK RT03 ...
mi.eng.cam.ac.uk/research/projects/EARS/pubs/jia_rt03s.pdf23 Jun 2003: Automatic Segmentation. Segmentation Diarisation CER (%)MS FA Tot. Manual 1.2 24.5 25.6 49.8Automatic 3.7 8.3 11.9 50.8. -
EFFICIENT USE OF END-TO-END DATA IN SPOKEN LANGUAGE PROCESSING ...
mi.eng.cam.ac.uk/~mjfg/ALTA/publications/lu_ICASSP2021.pdf22 Jul 2021: German translations are kept true-cased. Corpus Task #Sentences. TED-LIUM3 [23] ASR 268kWMT17-P En-De [24] MT 400kMuST-C En-De [25] ST 229k. ... E2E - - 19.29Casc 9.60 16.56 20.54EP 7.97 18.84 22.56. EP-J 9.60 16.24 23.25. -
Ongoing Experiments with Fisher Data Ricky Chan, Gunnar Evermann ...
mi.eng.cam.ac.uk/research/projects/EARS/pubs/chan_stthomas03.pdf10 Dec 2003: 5MPE fisher3896 (520h) MPE 26.4 30.5 22.1 28.3 24.6MPE fisher3896h5 (880h) MPE 25.7 29.9 21.3 27.4 24.1. ... 3 25.0 22.8fisher3896 LM03Fi 23.1 27.0 18.9 24.6 21.6fisher3896h5 LM03Fi 22.7 26.6 18.5 24.2 21.1. -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/segdisc_2012.pdf19 Oct 2012: α̂(r) = argminα. {. F(. α, w(r), O(r); α̂(r1))}. (15). Conditional Maximum Likelihood[24]:. ... . . . . . . . vj V (24). whereV is the vocabulary of segment identities. -
Combining Tandem and Hybrid Systems for Improved Speech…
mi.eng.cam.ac.uk/~mjfg/interspeech14-rath.pdf23 Jun 2014: A speaker adaptive training (SAT) system using global con-strained maximum likelihood linear regression (CMLLR) ata speaker level [24] was then constructed, followed by bothMinimum Phone Error (MPE) [25] and ... 7, no. 3, pp. 272–281, May 1999. [24] M. -
PROC IEEE, VOL. 101, NO. 5, 1160-1179, 2013 1 ...
mi.eng.cam.ac.uk/~sjy/papers/ygtw13.pdf20 Feb 2018: A similar approach is takenin [24], except that both slot values and their complements areused to build a frame. ... R(θ) = 1N. Nn=1. Tn1t=0. θ log π(ânt |bnt ,θ)Q(bnt , ânt ) (24). -
Low-Resource Speech Recognition andKeyword-Spotting M.J.F. Gales,…
mi.eng.cam.ac.uk/~mjfg/BABEL/SPECOM2017_paper.pdf11 Feb 2020: These approachesinclude data augmentation [23,5,14]; the use of web data [20]; extensive systemcombination [31]; and the use of multiple languages [4,24]. ... In: Proc InterSpeech (2014). 24. Ragni, A., Dakin, E., Chen, X., Gales, M.J., Knill, K.M.: Multi -
4F10: Deep Learning
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect6.pdf8 Nov 2016: represents element-wise multiplication between vectors. 24/68. Long-Short Term Memory Networks (reference) [13, 10]. ... Ẽ (θ[τ]) = E (θ[τ]) νw[τ]. 50/68. Dropout [24]. Input. xd. -
A HYBRID DISPLACEMENTESTIMATION METHOD FOR ULTRASONIC ELASTICITY…
mi.eng.cam.ac.uk/reports/svr-ftp/chen_tr615.pdf13 Nov 2008: IEEE Transactions on MedicalImaging, 23:153–163, 2004. [24] A. Pesavento, C. Perrey, M. ... Ultrasound in Medicine and Biology, 24:427–441, 1998. [31] R. Zahiri-Azar and S. -
SSVM_LVCSR.dvi
mi.eng.cam.ac.uk/~mjfg/sxz_TASLP12.pdf19 Oct 2012: Flat Direct Model [11] SCRF [23] or CAug [13]. M3N [18] orLarge Margin (Multi-Class) SVM [24]. ... Reformulate equation (24) in the form of (17). 1. 2||ᾱ||22+. C. -
system.dvi
mi.eng.cam.ac.uk/~sjy/papers/heyo06.pdf20 Feb 2018: 24. 0.6 0.7 0.8 0.9 10.5. 0.6. 0.7. 0.8. 0.9. 1. -
Kate Knill ALTA Institute, Cambridge University Engineering…
mi.eng.cam.ac.uk/~kmk/presentations/Knill_ISCSLP_Keynote_2022.pdf19 Jun 2023: 24. GEC. 25. seq2seq GEC. Treat GEC as a machine translation problem i.e. -
IEEE TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING, VOL. ...
mi.eng.cam.ac.uk/~sjy/papers/scyo09.pdf20 Feb 2018: and for the transition into. Inspired byagenda-based approaches to dialogue management [23], [24]the user state is factored into an agenda and a goal suchthat where consists of constraints. ... 24)The two update steps can be treated separately and -
KERNELIZED LOG LINEAR MODELS FOR CONTINUOUS SPEECH RECOGNITION…
mi.eng.cam.ac.uk/~mjfg/Kernel/sxz20_icassp13.pdf27 Mar 2013: 8 1.710 4.4 3.6 3.5 3.3 3.205 11.2 9.2 8.9 8.8 8.400 29.6 25.1 24.9 24.1 ... Speech Lang.,vol. 24, no. 4, pp. 648–662, 2010. [21] H. Liao and M. -
Part IA Computing CourseTutorial Guide to C++ Programming Roberto ...
mi.eng.cam.ac.uk/~cipolla/resource/tutorial.pdf12 Apr 2022: 24. // testing for real solutions to a quadraticd = bb - 4ac;if(d >= 0.0){. // -
THE CAMBRIDGE UNIVERSITY SPOKEN DOCUMENT RETRIEVAL SYSTEM S.E.…
mi.eng.cam.ac.uk/reports/svr-ftp/johnson_icassp99.pdf8 Mar 2000: This system gave a word error rate of 24.1% on the TREC-6 test data and ran in about 50 times real time. ... Recogniser WER stop map stemIBM Baseline 50.0 47.5 47.2 44.3Sheffield 39.8 37.6 37.1 34.6HTK-1 28.6 24.9 24.7 -
JournalPaperDRAFTV0.20
mi.eng.cam.ac.uk/~sjy/papers/wipy05a.pdf20 Feb 2018: J. D. Williams, P. Poupart, S. Young 24 March 2005. University of Cambridge, Dept. ... 24. not “brittle” – i.e., they do not fail catastrophically as the actual value of errp deviates from that used in training. -
lect4.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect4.pdf10 Nov 2015: the KL divergence given above. 24 Engineering Part IIB: Module 4F10 Statistical Pattern Processing. -
Knill_IS14_final.dvi
mi.eng.cam.ac.uk/~mjfg/interspeech14-knill.pdf23 Jun 2014: domised at the frame level across all the languages [24, 5]. ... 29, no. 6, pp. 82–97, Nov 2012. [24] J.-T. Huang et al., “Cross-language knowledge transfer using mul-tilingual deep neural network with shared hidden layers,” in Proc.ICASSP, -
Deep Learning for Automatic Assessment and Feedback of Spoken English
mi.eng.cam.ac.uk/~mjfg/thesis_kk492.pdf17 Nov 2021: Deep Learning for AutomaticAssessment and Feedback of Spoken. English. Konstantinos Kyriakopoulos. Supervisor: Prof. Mark J.F. Gales. Department of EngineeringUniversity of Cambridge. This thesis is submitted for the degree of Doctor of Philosophy. -
Natural Actor and Belief Critic: Reinforcementalgorithm for learning…
mi.eng.cam.ac.uk/~sjy/papers/juty11.pdf20 Feb 2018: wTα,i,wTθ,i]T C. n {1,. ,N}Parameter update:. 24: αi1 αi1 βαwα,i25: θi1 θi1 βθwθ,i26: end for. -
Reinforcement learning for parameter estimation in statistical spoken …
mi.eng.cam.ac.uk/~sjy/papers/juty12.pdf20 Feb 2018: Given the 8 subgoals (excluding the “name” subgoal) in the domain, there are 24 slot-level summary actions in total.The global summary actions implement other actions which cannot be associated ... wTα,i, wTθ,i]T C n {1,. , N}. Parameter update:24: -
Applying Deep Learning in Non-native Spoken English Assessment
mi.eng.cam.ac.uk/~kmk/presentations/APSIPA2019_Knill_Keynote.pdf21 Nov 2019: 1.0 indicates within one CEFR grade-level. 24/45. Assessment System Performance. • ... 1.0 indicates within one CEFR grade-level. 24/45. Performance Analysis. 25/45. -
Multi-Language Neural Network Language Models
mi.eng.cam.ac.uk/~mjfg/interspeech16_MLNNLMs.pdf26 Sep 2016: Data-based schemes instead. make use of data to initialise [27], train [20, 23] or adapt [24] the. ... The amount. of training data in VLLP conditions is 31,959 and 24,703 words. -
A Practical Method for Estimation of PointLight-Sour ces Martin ...
mi.eng.cam.ac.uk/reports/svr-ftp/weber_bmvc2001.pdf29 Nov 2003: Object View! normalised 24 2 4 H H #cube. Z n @ D$L Z¡ LD ' n$ $ n[ D[ n[ ' D ¡¢ J n D Z n D @ n L 3LD Z ... variousimages! normalised is expectedto beapproximatelyone. Note that the variationsin. H H are much larger then variationsin 24 2 -
Correction of Probe Pressure Artifactsin Freehand 3D Ultrasound — ...
mi.eng.cam.ac.uk/reports/svr-ftp/treece_tr411.pdf27 Apr 2001: Ultrasound in Medicine and Biology, 24(6):855–869, 1998. [6] W. L. Smith and A. ... Parker. Multilevel and motion model-based ultrasonicspeckle tracking algorithms. Ultrasound in Medicine and Biology, 24(3):427–441, 1998. -
SPEECH RECOGNITION AND KEYWORD SPOTTING FOR LOW RESOURCELANGUAGES:…
mi.eng.cam.ac.uk/~mjfg/sltu14_mjfg.pdf6 Jun 2014: This is effectively anunsupervised acoustic model training process [24, 25]. For all experiments the core ASR toolkit, used for acousticfeature generation, clustering, decoding and GMM-based acousticmodel training, was an extended ... 7319–7323, IEEE, -
lect8.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect8_pres.pdf20 Nov 2015: 34. 24 Engineering Part IIB: 4F10 Statistical Pattern Processing. Feature-Space Example. ... 34. 24 Engineering Part IIB: 4F10 Statistical Pattern Processing. Dimensionality of the Feature Space. -
PREDICTIVE LINEAR TRANSFORMS FOR NOISE ROBUST SPEECH RECOGNITION…
mi.eng.cam.ac.uk/~mjfg/gales_ASRU07.pdf15 Aug 2007: r)b. ”A(r)T. ”(24). 4. Goto 2 unless converged, or maximum number of iterations.It is possible to stop the estimation at various stages. -
A NEW METHOD FOR THEACQUISITION OF ULTRASONIC STRAIN IMAGE ...
mi.eng.cam.ac.uk/reports/svr-ftp/housden_tr656.pdf10 Aug 2010: The individual image frames arelocated in space by a position sensor attached to the probe [16] or, in the case of intravascularimaging, by a continuous pullback method [24]. ... Communications of theACM, 24(6):381–395, June 1981. [10] T. G. Fisher, T. -
SENSORLESS FREEHAND 3DULTRASOUND IN REAL TISSUE: SPECKLE…
mi.eng.cam.ac.uk/reports/svr-ftp/gee_tr510.pdf10 Jan 2005: Nelson and D. H. Pretorius. Three-dimensional ultrasound imaging. Ultrasound inMedicine and Biology, 24(9):1243–1270, 1998. ... Pattern Recognition Letters, 24(4–5):705–713, February 2003. [13] G. Treece, R. Prager, and Gee. -
Semantic Specialisation of Distributional Word Vector Spaces…
mi.eng.cam.ac.uk/~sjy/papers/mvol17.pdf20 Feb 2018: English 640 5 246 11 356 24 196 9German - - 135 2 277 13 175 6Italian - - - - 159 7 220 11. ... Word Vectors English German Italian RussianMonolingual Distributional Vectors 0.32 0.28 0.36 0.38COUNTER-FITTING: Mono-Syn 0.45 0.24 0.29 0.46COUNTER-FITTING: -
DESIGN OF FAST LVCSR SYSTEMS G. Evermann & P.C. ...
mi.eng.cam.ac.uk/reports/svr-ftp/evermann_asru2003.pdf23 Sep 2003: System A B C D23.0 23.6 23.4 24.8. A 23.1 22.6 22.7B 22.9 23.3C 22.8. ... full ’02 19.8 24.3 27.0 23.910x ’02 22.3 27.7 31.0 27.2 14%. -
The 1998 HTK Broadcast News Transcription System:Development and…
mi.eng.cam.ac.uk/reports/svr-ftp/woodland_darpa99.pdf8 Mar 2000: Y 14.2 8.0 15.4 20.3 16.5 14.0 16.6 24.6. ... ROVER fgintcat 4/Y1/N 13.8 7.8 15.1 20.1 15.8 13.6 16.6 24.1. -
paper.dvi
mi.eng.cam.ac.uk/reports/svr-ftp/liu_asru2003.pdf25 Sep 2003: j (τ ) = γnum. j (τ ) γden. j (τ ) Cp(O, Sj,τ |Θ̃) (24). ... For more details see [15]. Two globalcomplexity attributes of a single transform HLDA system wereoptimized: the number of Gaussian components per state from therange {12, 16, 24}; the -
The Development of the Cambridge UniversityRT-04 Diarisation System…
mi.eng.cam.ac.uk/research/projects/EARS/pubs/tranter_rt04.pdf15 Feb 2005: 18.5. 19. 19.5. 20. mea. n D. ER. on. all. 24 d. ... RT-04f RT-04f RT-04f 23.86 24.12† Official evaluation submission. • New segmenter and clusterer resulted in 14% absolute drop in primary DER. • -
Kim et al.: 2003 CU-HTK BN-E Systems 2003 CU-HTK ...
mi.eng.cam.ac.uk/research/projects/EARS/pubs/kim_rt03s.pdf24 Jul 2003: Lattice-Regen 14.4 8.5 15.1 17.7 16.9 14.6 21.3 24.4 14.5 13.8. -
DEVELOPMENT OF THE CU-HTK 2004 BROADCAST NEWS TRANSCRIPTION SYSTEMS…
mi.eng.cam.ac.uk/reports/svr-ftp/kim_icassp05.pdf1 Apr 2005: Using improved acoustic/language models andby combining systems with different segmentations, the 10RTsystem gave on average a 24% relative reduction in WER over theRT03 system. -
williams2006IEEE.dvi
mi.eng.cam.ac.uk/~sjy/papers/wiyo07b.pdf20 Feb 2018: Similarly, attempting to exploit the factoredform of the SDS-POMDP in optimization (using e.g., [24],[25]) is unlikely to succeed since most of the growth isdue to one component (the -
EXPERIMENTS IN BROADCAST NEWS TRANSCRIPTION P.C. Woodland, T. Hain,…
mi.eng.cam.ac.uk/reports/svr-ftp/woodland_icassp98.pdf10 Apr 2000: M S TM 82.11 17.89 0.00N 15.27 84.22 0.51S 0.56 98.24 1.20T 0.00 1.19 98.81. ... 1F3 36.4 32.7 35.3 33.9F4 28.6 25.0 25.4 24.4F5 28.6 23.8 27.1 26.5FX 58.5 55.2 56.8 -
lect5.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect5_pres.pdf10 Nov 2015: g" plane. solution region. aa. 24. 10 Engineering Part IIB: Module 4F10 Statistical Pattern Processing. -
unsup.eps
mi.eng.cam.ac.uk/~mjfg/interspeech14-ragni.pdf23 Jun 2014: Thusmany of the waveform production issues [23] are not rele-vant. Furthermore, these schemes permit model-based adap-tation/compensation approaches [24] to be used for synthesis-ing data with target ... 51, pp. 1039–1064, 2009. [24] M. J. F. Gales, -
The CUHTK-Entropic 10xRT Broadcast News Transcription SystemJ.J.…
mi.eng.cam.ac.uk/reports/svr-ftp/odell_darpa99.pdf8 Mar 2000: 22. 23. 24. 25. 26. 27. 28. 29. 30. 0 5 10 15 20 25. ... Comparing Table 4 with Table 3the advantage of including two passes can clearly be seen: onBNeval98 the error rate was reduced by 24% while the overallcomputation was increased by about a -
Hand PoseEstimation Using Hierar chical Detection B. Stenger��� , ...
mi.eng.cam.ac.uk/reports/svr-ftp/stenger_hci04.pdf8 Aug 2005: Both, centretemplateandmarginalizedtemplateshow betterclassificationperfor-mancethanthe trainedclassifier, in particular in the high detection range.At de-tectionratesof 0.99 thefalsepositive ratefor thecentretemplateis 0.24, wherasitis 0.64for
Refine your results
Date
- 455 2018
- 124 2005
- 40 2000
- 33 2006
- 29 2003
- 26 2023
- 25 2022
- 23 Past year
- 21 2004
- 20 2014
- 20 Uncertain
- 18 2007
- 17 2019
- 17 2016
- 17 2015
- 15 2009
- 14 2010
- 13 Past 6 months
- 13 2008
- 13 2024
- 12 2017
- 12 2001
- 11 2013
- 10 2021
- 10 2020
- 9 2012
- 7 2002
- 6 2011
- 6 Past 3 months
- 6 1999
- 1 Past week
- 1 Past fortnight
- 1 Past month
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.