Search
Search Funnelback University
151 -
200 of
1,003
search results for TALK:PC53 20 |u:mi.eng.cam.ac.uk
where 0
match all words and 1,003
match some words.
Results that match 1 of 2 words
-
Reward-Balancing for Statistical Spoken Dialogue Systems…
mi.eng.cam.ac.uk/~sjy/papers/ubcy17.pdf20 Feb 2018: 0. 5. 10. 15. 20. 0.0. 0.2. 0.4. 0.6. 0.8. 1.0. ... domains comparing the baseline (rws = 20, w = (0.5,0.5)) with the optimised balance. -
The Cambridge University Multimedia Document Retrieval Demo System
mi.eng.cam.ac.uk/reports/full_html/tuerk_sigir00demo.html/14 Aug 2000: On the Internet audio used here the run-time is increased due to reduced audio quality but the general level of transcription accuracy remains high at approximately 20% word error rate -
User study of the Bayesian Update of Dialogue State ...
mi.eng.cam.ac.uk/~sjy/papers/tgkm0820 Feb 2018: 2) was chosen. 0 20 40 60 80 1000. 0.2. 0.4. ... 0.6. 0.8. 1MDPHDC. 0 20 40 60 80 1000. 0.2. 0.4. -
CUHTK Brodacast News English Recipe
mi.eng.cam.ac.uk/~mjfg/private/CUHTK-BNE-Recipe/14 Jun 2004: DARPA Broadcast News Recognition Workshops. Last modified: Fri Oct 11 19:55:20 BST 2002. -
Complementary System Combination andGeneration for ASR Mark Gales 20…
mi.eng.cam.ac.uk/~mjfg/mjfg_tcstar.pdf5 Jul 2006: factorial HMMs [19], loosely coupled models [20], system combination [16]– synchronous: likelihoods combined at the state level. ... 0. 5. 10. 15. 20. 25. 30. 35. 40. 0 1000 2000 3000 4000 5000 6000 7000 8000 9000. % -
Deep Learning for Speech Recognition
mi.eng.cam.ac.uk/~mjfg/LxMLS17.pdf29 Nov 2017: 39/57. Example “Generative” Acoustic Model [20]. understand the CLDNN architecture are presented in Section 4. ... Learning Internal Representations by Error Propagation, pp.318–362. [20] T. N. Sainath, O. -
34 1 2 Localaffine deformation Total affinedeformation Robot…
mi.eng.cam.ac.uk/~cipolla/publications/article/2000-IJCV-visual-servo.pdf13 Mar 2018: 000000. 0"/20. 0. 00#$0. =. 0. $#$0000. 00. ( not! )0#"/200. ... 0 5 10 15 20 25 30 35. "fourier.txt". 0. 0.02. -
Deep Neural Network Approach for theDialog State Tracking Challenge…
mi.eng.cam.ac.uk/~sjy/papers/hety13.pdf20 Feb 2018: Structure[] 0.6751 0.8074 0.5658. [50] 0.6679 0.8046 0.5450[20] 0.6656 0.8060 0.5394. ... Training used T =10, the full feature set,a [20, 10, 2] hidden structure and the shared ini-tialisation training method. -
INDICATOR VARIABLE DEPENDENT OUTPUT PROBABILITY MODELLING…
mi.eng.cam.ac.uk/~sjy/papers/tuyo01.pdf20 Feb 2018: avg. F0 F1 F2 F3 F4 F5 FX12mix base 21.8 12.4 20.5 31.5 31.5 24.3 27.5 43.924mix base 21.3 12.2 ... relabelled training datasum 21.4 12.5 20.1 30.2 31.2 24.7 25.1 42.2full 21.4 12.5 20.1 30.6 31.0 24.6 -
Machine Intelligence Laboratory
mi.eng.cam.ac.uk/Main/GMT_VolMorphDefault is 20. -p {points}. sampling points. {points} sets the minimum number of points, in any direction, used to sample the initial and final files. ... The default is 20. -r {resolution}. triangulation resolution. {resolution} defines the resolution -
yeyo05b.dvi
mi.eng.cam.ac.uk/~sjy/papers/yeyo05b.pdf20 Feb 2018: Native: Chinese sentences 11-20 and 31-40 spoken by all 20Chinese speakers, 400 utterances in total. ... NonNativeAB: Chinese sentences 11-20 and 31-40 spoken byall 20 English speakers, 400 utterances in total. -
ICSLPDataCollection-10
mi.eng.cam.ac.uk/~sjy/papers/wiyo04b.pdf20 Feb 2018: 0. 5. 10. 15. 20. 25. 30. 35. Hi Med Low NoneWER Target. ... 0%. 20%. 40%. 60%. 80%. 100%. REPAIR ASKQ REPAIR ASKQ. Hi MedStrategy & WER Target. % -
a) Rotation b) Isotropic Expansion c) Pure Shear 0 ...
mi.eng.cam.ac.uk/~cipolla/publications/inproceedings/1996-BMVC-Cross.pdf13 Mar 2018: 17. 17.5. 18. 18.5. 19. 19.5. 20. 20.5. Sample Number. Tim. -
Spoken Document Retrieval for TREC-7 at Cambridge University
mi.eng.cam.ac.uk/reports/full_html/johnson_trec7.html/30 Mar 2000: 20.4. 5.9. 4.8. 31.0. NIST/CMU base1. 72.1. 22.6. 5.3. 6.7. 34.6. ... 0.4556. 0.4408. 0.5739. tune. 0.4903. 0.4639. 0.6000. 0.4567. 0.4493. 0.5652. Table 20: Cumulative Improvements on TREC-7 without wp. -
k-Nearest Neighbor Monte-Carlo Control Algorithmfor POMDP-based…
mi.eng.cam.ac.uk/~sjy/papers/lgjk09.pdf20 Feb 2018: 0 10 20 30 40 50. Succ. essf. ul C. ompl. ... 7. 8. 9. 10. 11. 12. 13. 14. 0 10 20 30 40 50. -
Intonation Modelling and Adaptation for Emotional Prosody Generation…
mi.eng.cam.ac.uk/~sjy/papers/inyo0520 Feb 2018: When analyzed on a per listener basis, the distribution of preferences in partone (happy) suggest strong bimodality: eight of the fourteen listeners had a meanidentification rate of 5 out of 20, ... Proc. of Speech Communication (20 and 03) vol.40 -
Toward Multi-domain Language Generation usingRecurrent Neural…
mi.eng.cam.ac.uk/~sjy/papers/wgmr15.pdf20 Feb 2018: During decoding, we over-generated 20 utterances and selected the top 5 realisations for each DAaccording to the following reranking criteria,. ... Proceedings of the IEEE,1990. [20] Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. -
Reward Shaping with Recurrent Neural Networks for Speeding upOn-Line…
mi.eng.cam.ac.uk/~sjy/papers/svgm15.pdf20 Feb 2018: Per turn reward was set to -1and final reward 20 for task success else 0. ... In Proc of the IEEE, vol-ume 99, pages 1–20. A Training Constraint/Loss Function. -
lect2.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect2.pdf25 Oct 2011: 10. 20. -10. 0. 10. 20. 0. 0.005. 0.01. P. -10. ... 0. 10. 20. 0. 0.005. 0.01. p. -20. -10. 0. 10. -
Emotion Conversion using F0 Segment Selection Zeynep Inanoglu, Steve…
mi.eng.cam.ac.uk/~sjy/papers/inyo08.pdf20 Feb 2018: 20% 19.3% 60.7%. 21.3% 47.3% 31.3%. Figure 2: Preference scores for each F0 prediction method andemotion. ... Angry Surprised Sad Can’t decideAngry 99.3% 0.7% 0% 0%. Surprised 20.0% 66.0% 0% 14.0%Sad 0.7% 0% 96.0% 3.3%. -
lect14
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect14.2up.pdf10 Nov 2015: 0.1 0.2 0.5 1 2 5 10 20 40. 0.1. 0.2. ... 0.5. 1. 2. 5. 10. 20. 40. False Alarm probability (in %). Miss. pro. babi. lity (i. n %. ). GMM SVM. MAP (τ)EER. -
Engineering Tripos Part IIB FOURTH YEAR Paper 4F10: Statistical ...
mi.eng.cam.ac.uk/~mjfg/local/4F10/examples2.pdf10 Nov 2016: ω1 :. [11. ] [22. ] [20. ]. ω2 :. [00. ] -
Using Wizard-of-Oz simulations to bootstrap…
mi.eng.cam.ac.uk/~sjy/papers/wiyo03.pdf20 Feb 2018: 6 Discussion. The majority of the data collected was “usable”: although 26.7% of turns were excluded, 20.5% of these were due to a well-defined task not under study ... 20.0%. 30.0%. 40.0%. 50.0%. 60.0%. 70.0%. 80.0%. 0 500 1000 1500 2000 -
acl2010.dvi
mi.eng.cam.ac.uk/~sjy/papers/gjkm10.pdf20 Feb 2018: 1. 20 60 100 140 180 220 260 300 340 380 420 460 500 540 580 620. ... 50. 45. 40. 35. 30. 25. 20. 15. 10. 5. 0. -
ERROR SIMULATION FOR TRAINING STATISTICAL DIALOGUE SYSTEMS Jost…
mi.eng.cam.ac.uk/~sjy/papers/scty07b20 Feb 2018: A computationally less expensive word-level error simulationmethod for training statistical dialogue systems has been suggestedby [20]. ... 20] O. Pietquin and T. Dutoit, “A probabilistic framework for dia-log simulation and optimal strategy -
ON-LINE POLICY OPTIMISATION OF BAYESIAN SPOKEN DIALOGUE SYSTEMS…
mi.eng.cam.ac.uk/~sjy/papers/gbhk13.pdf20 Feb 2018: Thismethod can be elegantly incorporated into the on-line GP-Sarsa al-gorithm [20]. ... The reward function was set to give a reward of 20 for success-ful dialogues, zero otherwise. -
WILLIAM JOSEPH BYRNE III Department of Engineering 16 Water ...
mi.eng.cam.ac.uk/~wjb31/byrnecv.pdf9 Nov 2023: 20. V. Goel, Ph.D. in Biomedical Engineering, 2001Thesis: Minimum Bayes-Risk Automatic Speech Recognition. ... Journal of Machine Learning Research, (6):2049–2073 (25 pages), December 2005. [20] V. -
The Cambridge University Multimedia Document RetrievalDemo System…
mi.eng.cam.ac.uk/reports/svr-ftp/tuerk_sigir00demo.pdf3 May 2000: On the Internet audio used herethe run-time is increased due to reduced audio qualitybut the general level of transcription accuracy remainshigh at approximately 20% word error rate on NPR data. -
POLICY COMMITTEE FOR ADAPTATION IN MULTI-DOMAIN SPOKEN…
mi.eng.cam.ac.uk/~sjy/papers/gmsv15.pdf20 Feb 2018: The reward function allocates 1 at each turnto encourage shorter dialogues, plus 20 at the end of each suc-cessful dialogue. ... 20] S Keizer, M Gašić, F Jurčı́ček, F Mairesse, B Thomson,K Yu, and S Young, “Parameter estimation for -
Dialogue Context Sensitive Speech Synthesis using Factorized Decision …
mi.eng.cam.ac.uk/~sjy/papers/tgbh14.pdf20 Feb 2018: A set of 20 previously collected dialogues were randomlyselected for the restaurant domain. ... 90, no. 9, pp.1406–1413, 2007. [20] K. Yu, H. Zen, F. -
Evaluating semantic-level confidence scores with multiple hypotheses…
mi.eng.cam.ac.uk/~sjy/papers/tygk08.pdf20 Feb 2018: 5 19.2 12.2 0.212 1.10335.3 InfEv 100 20.6 7.9 — 0.941. ... 0 10 20 30 40 500. 1. 2. 3. 4. Oracle Error Rate. -
The Cambridge Multimedia Document Retrieval (MDR) Project : Summary…
mi.eng.cam.ac.uk/reports/full_html/sparckjones_cltr517.html/10 Oct 2001: 20 40.90 32.97 38.65 41.45 39.05 36.22 BE DBRF3 (X) 65.42 65.04 45.12 42.50. ... t=200 r=20, tf=0.5 (same as T8) X: For Trec-8 -> Nov 96 to Jan 98 (pre-trec-8) = 60,000 stories (129614 terms). -
Learning from Real Users: Rating Dialogue Success with Neural ...
mi.eng.cam.ac.uk/~sjy/papers/svgk15.pdf20 Feb 2018: 0. This hard label is used to determine whether to give afinal reward of 20 during policy learning, as per Eqn. ... All policies are trained by GP-SARSA [4] andthe summary action space contains 20 actions. -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/ragni_ASRU11.pdf20 Dec 2011: The standardapproach with generative models is to usestate-level phoneticdecision trees to cluster phonetically similar states together[20]. ... Interspeech, 2011. [20] S. Young, J. Odell, and P. Woodland, “Tree-based state tying for highaccuracy -
Domain-independent User Satisfaction Reward Estimation forDialogue…
mi.eng.cam.ac.uk/~sjy/papers/ubcm17.pdf20 Feb 2018: Sec. 2. the reward [20, 21, 22, 23] by using the PARADISE frame-work [24]. ... IEEE, 2015, pp. 763–770. [20] M. Walker, J. C. Fromer, and S. -
draft21.dvi
mi.eng.cam.ac.uk/~mjfg/ASRU13.pdf7 Nov 2013: Pashto 104 12 44 20 28 20. Turkish 105 10 42 26 19. ... and Language Processing, vol. 20, no. 3, pp. 875–887, 2012. [31] S. -
Cross-Lingual Spoken Language Understanding from Unaligned Data…
mi.eng.cam.ac.uk/~sjy/papers/lemy10.pdf20 Feb 2018: In order to minimise the amount of domain-specific datain the target language, we rely on various external resources.First, we use the Europarl corpus [9], which contains over 20. -
williams2006aaai.dvi
mi.eng.cam.ac.uk/~sjy/papers/wiyo06b.pdf20 Feb 2018: 10. 15. 20. 0.05. 0.10. 0.15. 0.20. 0.25. 0.30. 0.35. 0.40. ... 0.10. 0.15. 0.20. 0.25. 0.30. 0.35. 0.40. 0.45. 0.50. 0.55. 0.60. -
INVESTIGATION OF ACOUSTIC MODELING TECHNIQUES FOR LVCSR SYSTEMS X. ...
mi.eng.cam.ac.uk/reports/svr-ftp/liu_icassp2005.pdf19 May 2005: The finalerror rates were 20.5% on eval03 and 16.9% on dev04. ... P3aP3c 23.9 16.8 20.5 16.9. Table 1. CTS 10xRT system baseline performance. -
3_2_ransac
mi.eng.cam.ac.uk/~cipolla/lectures/4F12/Slides/4F12-ImageStitching.pdf27 Oct 2020: 20. 2-view Rotation Estimation• Remove outliers, can now solve for R using least squares. -
Machine Intelligence Laboratory
mi.eng.cam.ac.uk/Main/GMT_IsoSurfMain > GMT_IsoSurf. Dr Graham Treece, Department of Engineering. IsoSurf. Note that the functionality in this software is now available (and in many ways improved) in Stradview, which is much more powerful and has a much nicer interface. The only -
Abstract for johnson_trec8
mi.eng.cam.ac.uk/reports/abstracts/johnson_trec8.html27 Jul 2020: The system gave an overall word error rate of 20.5% on the 10 hour scored subset of the corpus, the lowest in the track. -
Abstract for johnson_trec9
mi.eng.cam.ac.uk/reports/abstracts/johnson_trec9.html27 Jul 2020: The CU-HTK transcriptions from TREC-8 with Word Error Rate (WER) of 20.5% were used in conjunction with stopping, Porter stemming, Okapi-style weighting and query expansion using a -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/ragni_ICASSP11.pdf11 Mar 2011: System CritTest set. AvgA B C D. HMM (VAT) ML 8.6 13.8 12.0 20.1 16.0CAUG MPE 7.5 13.0 10.9 19.4 15.3. ... SNR is randomly chosen bewteen 5 and 15 dBfor testdata and between 10 and 20 dB for multi-style training data. -
JOINT MODELLING OF VOICING LABEL AND CONTINUOUS F0 FOR ...
mi.eng.cam.ac.uk/~sjy/papers/yuyo11a.pdf20 Feb 2018: Toreduce the noise introduced by forced choices, the 10 wave file pairswere duplicated and the order of the two systems were swapped.The final 20 wave file pairs were then shuffled -
Machine Intelligence Laboratory
mi.eng.cam.ac.uk/Main/GMTPublicationsMedical Image Analysis, Vol. 20, No. 1, pp. 249-264, January 2015. ... Proceedings of Medical Image Understanding and Analysis, pp. 17-20, Oxford, July 1999. -
Clinically Practical Freehand Three-Dimensional Ultrasound
mi.eng.cam.ac.uk/research/projects/cp3dus/22 Jul 2010: We have developed novel reconstruction algorithms to relax this constraint [1, 2, 15, 20, 23, 31, 33]. ... 20] R.J. Housden, A.H. Gee, R.W. Prager and G.M. Treece. Sensorless freehand 3D ultrasound for non-monotonic and intersecting frames. -
MultiMedia Document Retrieval (1997-2000)
mi.eng.cam.ac.uk/research/projects/Multimedia_Document_Retrieval/7 Oct 2001: The final WER on the 10 hour scored subset of the TREC-8 SDR corpus was 20.6% (the lowest in the evaluation). -
Machine Intelligence Laboratory
mi.eng.cam.ac.uk/Main/AHGPublicationsMedical Image Analysis, 20(1):249-264, February 2015. K.E.S. Poole, G.M. Treece, A.H. ... In Proceedings of Medical Image Understanding and Analysis, pages 17-20, Oxford, UK, 1999. -
A Method for Direct Audio Search with Applications to Indexing and…
mi.eng.cam.ac.uk/reports/full_html/johnson_icassp00.html/16 Jun 2000: However the transcription systems used (see e.g. [6]) typically run rather slower than real-time and have a word error rate of over 20%. ... 967. 5480s=61.0%. 114s=0.5%. window, 20@4s. 284. 4598s=51.2%. 122s=0.5%. Automatic segs. 38.
Refine your results
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.