Search
Search Funnelback University
201 -
250 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
-
From Discontinuous To Continuous F0 Modelling In HMM-based…
mi.eng.cam.ac.uk/~sjy/papers/yuty10.pdf20 Feb 2018: GTD 23.44 7.06 20.25 8.10CF 12.58 7.29 11.90 8.43. Table 1: Objective comparison between MSDHMM extensionsand CF-HMM. ... The final 20 waves were then shuffled and presentedto the listeners in random order. -
lect3_pres.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect3_pres.pdf10 Nov 2015: 0.05. 0.1. 0.15. 0.2. 0.25. 0.3. 0.35. 20. 4 Engineering Part IIB: Module 4F10 Statistical Pattern Processing. ... 4. 2. 0. 2. 4. 6. 83 ITERATIONS. 20 Engineering Part IIB: Module 4F10 Statistical Pattern Processing. -
Context Adaptive Training with Factorized Decision Treesfor HMM-Based …
mi.eng.cam.ac.uk/~sjy/papers/yzmy10.pdf20 Feb 2018: Theperformance of emphasized word detection is shown in table 1. System # Det Rec (%) Pre (%) F -measureGMM 2.0 20.0 53.8 0.29MLLR 4.7 47.1 68.0 -
paper.dvi
mi.eng.cam.ac.uk/reports/svr-ftp/liu_icassp2004.pdf29 May 2004: Gauss 24.0 20.7 20.7 17.7 16.0WER (%) 35.3 35.1 35.2 35.3 35.5. ... Using the “opti-mal” structure, determined with α = 0 and 4 iterations of structureoptimization the error rate fell to 35.1% and the average numberof components per state was 20.7. -
Statistical User Simulation with a Hidden Agenda Jost Schatzmann ...
mi.eng.cam.ac.uk/~sjy/papers/scty07.pdf20 Feb 2018: Q(θ, θ(k1)) =. X. P (X, D|θ(k1)) log P (X, D|θ).(20). ... 50. -40. -30. -20. -10. 01 2 3 4 5 6 7 8 9 10. -
eps.dis.dur.testa.eps
mi.eng.cam.ac.uk/~mjfg/gales_ASRU09.pdf14 Sep 2010: SYN HMM 30 9.20 8.51 9.34 9.02HTS 5 8.41 8.03 8.70 8.38. ... Beyond HMM Workshop, December 2004. [20] H. AlDamarki, “Filter trees for noise robust small vocabulary speechrecognition,” M.Phil. -
THE 1998 HTK SYSTEM FOR TRANSCRIPTION OFCONVERSATIONAL TELEPHONE…
mi.eng.cam.ac.uk/reports/svr-ftp/hain_icassp99.pdf27 Sep 2000: The worderror rate obtained is almost 20% better than our 1997 system onthe development set. ... The eval97sub set was used forsystem development and consisted of 20 conversation sides fromSwbd-II and CHE. -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/sxz20_inter11.pdf19 Jan 2012: 0 50 100 150 200 2500. 20. 40. 60. 80. 100. ... Just optimising theinference alignment gave2.1% relative reduction. The overallgain from using the SSVM over the VTS-compensated HMMsystem was over 20%, though it should be noted that the SVMand SSVM -
Dialogue manager domain adaptation using Gaussian process…
mi.eng.cam.ac.uk/~sjy/papers/gmrs17.pdf20 Feb 2018: 20. Table 5: Selection of committee members for multi-policy Bayesian committee machinefor SFR domain. ... 250 SFR250 SFH250 L11 8.05 0.20 79.38 0.83 7.79 0.08. 6.8. -
Challenges for AI in Spoken Communication March 2017 Dr ...
mi.eng.cam.ac.uk/~kmk/presentations/CUED_Knill_20170328_enc.pdf22 Jan 2018: Unless otherwise indicated,the LSTM is unrolled for 20 time steps for training with truncatedbackpropagation through time (BPTT). ... The LSTM is then unrolled for 20timesteps, and thus consumes a larger context of 20 l. -
WORD-LEVEL EMPHASIS MODELLING IN HMM-BASED SPEECH SYNTHESIS K. Yu, ...
mi.eng.cam.ac.uk/~sjy/papers/yumy10.pdf20 Feb 2018: The final 20 samples were then shuffled andprovided to the listeners. -
yeyo06.dvi
mi.eng.cam.ac.uk/~sjy/papers/yeyo06.pdf20 Feb 2018: utterances. ML LSEpreference 48.3% 51.7%. 20 40 60 80 100 120 140 54. ... This filter ispopular in speech coding [20] and its more general use invoice conversion is discussed in [6]. -
PII: S0167-6393(99)00044-8
mi.eng.cam.ac.uk/~sjy/papers/wiyo00.pdf20 Feb 2018: Received 20 February 1998; received in revised form 26 November 1998; accepted 2 December 1998. ... Additionally, 20 sentences from a femaleSpanish speaker were recorded to serve as cali-bration sentences. -
Low-Resource Speech Recognition and Keyword-Spotting
mi.eng.cam.ac.uk/~mjfg/SPECOM_2017.pdf29 Nov 2017: 20/63. Phonetic vs Graphemic Performance. Language Id Script TER (%)Phon Grph CNCTok Pisin 207 Latin 40.6 41.1 39.4Kazakh 302 Cyrillic/Latin 53.5 52.7 51.5Telugu -
On-line Active Reward Learning for Policy Optimisationin Spoken…
mi.eng.cam.ac.uk/~sjy/papers/sgmb16.pdf20 Feb 2018: All policies were trained using the GP-SARSA al-gorithm and the summary action space of the RLpolicy contains 20 actions. ... 2013. Pomdp-basedstatistical spoken dialogue systems: a review. InProc of IEEE, volume 99, pages 1–20. -
EFFICIENT LATTICE RESCORING USINGRECURRENT NEURAL NETWORK LANGUAGE…
mi.eng.cam.ac.uk/~mjfg/xl207_ICASSP14a.pdf28 Apr 2014: 3. HISTORY CONTEXT CLUSTERING FOR RNNLMS. Efficient use of language models in speech recognizers [20, 19, 17]requires that the context dependent states representing different his-tories during search can be ... In common with the n-gram history based -
Applying Deep Learning in Non-native Spoken English Assessment
mi.eng.cam.ac.uk/~mjfg/ALTA/presentations/APSIPA2019_Knill.pdf21 Feb 2022: 20/45. Assessment: Gaussian Process [14, 16]. • Gaussian process• non-parametric model based on joint-Gaussian assumption. • ... 16-20, 2017, 2017, pp. -
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. -
A LANGUAGE SPACE REPRESENTATION FOR SPEECH RECOGNITION
mi.eng.cam.ac.uk/~mjfg/icassp15-ragni.pdf18 May 2015: Though for speaker adaptation this may be reason-able [20], for language adaptation this is expected to be suboptimalas different eigen-languages may have different acoustic realisationsof context-dependencies. ... Rep. CUED/F-INFENG/TR263,Cambridge -
Discriminative Classifiers with Generative Kernels for Noise Robust…
mi.eng.cam.ac.uk/~mjfg/gales_INTER08.pdf16 Jun 2008: from 20 to 5 dB,combined with the 4 different additive noise sources N1 to N4,subway, babble, car and exhibition hall. ... A set of 20 confusable digit-pairs were selected based onthe overall confusion matrix for the 16 noise conditions. -
Towards Learning Orientated Assessment for Non-native Learner Spoken…
mi.eng.cam.ac.uk/~kmk/presentations/ALTA_Sheffield_20190306.pdf8 Mar 2019: 400 hour BULATS training set. 20. AM LM % WER. -
Model Refinementfr om Planar Parallax A. R. Dick R. ...
mi.eng.cam.ac.uk/reports/svr-ftp/dick_bmvc99.pdf5 Dec 2003: 17] T. Vieville, C. Zeller, andL. Robert. Using collineationsto computemotion andstructurein an uncali-bratedimagesequence.International Journal of Computer Vision, 20(3):213–242,1996. -
Structure and Motion from Silhouettes Kwan-Yee K. Wong and ...
mi.eng.cam.ac.uk/reports/svr-ftp/wong_iccv01.pdf19 Apr 2001: Wewould also like to extend the system to recover surface re-flectance [8, 20], so as to produce photo-realistic 3D modelsunder different lighting conditions. ... IEEE Trans. on Pattern Analysis and Machine In-tell., 20(10):1091–1096, Oct 1998. -
Kate Knill ALTA Institute, Cambridge University Engineering…
mi.eng.cam.ac.uk/~kmk/presentations/Knill_UKSpeech2023_Keynote.pdf19 Jun 2023: 20. Spoken Language Assessment and Feedback Pipeline. Grader Score: 3.5Conf: 90%. -
EFFICIENT USE OF END-TO-END DATA IN SPOKEN LANGUAGE PROCESSING ...
mi.eng.cam.ac.uk/~mjfg/ALTA/publications/lu_ICASSP2021.pdf22 Jul 2021: Various models shown in Figure 1are discussed below.Cascade (Casc) The vanilla Casc model consists of an RNN-basedListen-attend-spell (LAS) [20] and a transformer-based NMT [10].In the ... E2E - - 19.29Casc 9.60 16.56 20.54EP 7.97 18.84 22.56. EP-J 9.60 -
Inverse Reinforcement Learning for Micro-Turn Management Dongho Kim,…
mi.eng.cam.ac.uk/~sjy/papers/kgbt14.pdf20 Feb 2018: 0.24. 0.22. 0.20. 0.18. 0.16. Log-. likel. ihoo. dpe. rmic. ro-tu. -
PREDICTIVE LINEAR TRANSFORMS FOR NOISE ROBUST SPEECH RECOGNITION…
mi.eng.cam.ac.uk/~mjfg/gales_ASRU07.pdf15 Aug 2007: Here. Σ(m)-1 = A(r)TΣ̃(m)-1diag A. (r) (20). and Σ̃(m)diag is a diagonal covariance matrix. ... System # iter SNR(dB)20 15 10 50 31.2 50.4 79.2 89.7. -
Context adaptive training with factorized decisiontrees for HMM-based …
mi.eng.cam.ac.uk/~sjy/papers/yzmy11.pdf20 Feb 2018: 20). where D is the data dimension, γ(θ) and Σ(θ) are the total occupancy and the. ... System # Det Rec (%) Pre (%) F -measure. GMM 2.0 20.0 53.8 0.29. -
Mark John Francis Gales Thesis, Book Chapters and Magazines ...
mi.eng.cam.ac.uk/~mjfg/publications_list.pdf11 Oct 2023: Joint Conference on NaturalLanguage Processing, AACL/IJCNLP 2022 - Volume 2: Short Papers, Online only, November 20-23, 2022 (Y. ... Speech and Signal Processing, ICASSP 2016,Shanghai, China, March 20-25, 2016, pp. -
Advances in Structural Metadata for RT-04 atCUED M. Tomalin ...
mi.eng.cam.ac.uk/research/projects/EARS/pubs/tomalin_rt04.pdf15 Feb 2005: 0/56.8 29.2/15.7/56.2. PFMctsrt04 cl40-tg 33.1/20.3/63.9 33.3/18.7/62.6 30.8/19.7/61.9. ... Cambridge University RT-04 workshop: November 2004 20. Tomalin and Woodland: Advances in Structural Metadata for RT-04 at CUED. -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/flego_icassp12.pdf13 Jun 2013: The same trends phone-. System A B C D avgVAT 8.1 13.5 11.6 20.5 15.98. -
williams2005continuous06
mi.eng.cam.ac.uk/~sjy/papers/wipy05c.pdf20 Feb 2018: 0.20. 0.25. 0.30. 0.35. 0.40. 0.45. 0.50. 0.55. 0.60. 0.65. perr. ... 0. 0.5. 1. 1.5. 2. 2.5. 0.00. 0.05. 0.10. 0.15. 0.20. -
AAAI Proceedings Template
mi.eng.cam.ac.uk/~sjy/papers/wipy05b.pdf20 Feb 2018: 0.05. 0.10. 0.15. 0.20. 0.25. 0.30. 0.35. 0.40. 0.45. 0.50. 0.55. ... 4. 6. 8. 10. 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 -
Infinite Support Vector Machines in Speech Recognition Jingzhou Yang, …
mi.eng.cam.ac.uk/~mjfg/yang_is2013.pdf13 Jun 2013: Thesampling process of all other parameters is similar to the meth-ods discussed in [14, 19, 20]. ... 9, no. 2, pp. 249–265, 2000. [20] Carl Edward Rasmussen and Zoubin Ghahramani, “Infinite mix-tures of Gaussian process experts,” in NIPS, 2001, pp. -
lect5.dvi
mi.eng.cam.ac.uk/~mjfg/local/4F10/lect5.pdf10 Nov 2015: 1. 1. x1. 1x2. 0.5NOR. (d) NOR operator. 20 Engineering Part IIB: Module 4F10 Statistical Pattern Processing. -
POMDP-based dialogue manager adaptation to extended domains M.…
mi.eng.cam.ac.uk/~sjy/papers/gbhk13a.pdf20 Feb 2018: The reward function was set to give a rewardof 20 for successful dialogues, zero otherwise. ... 20. 15. 10. 5. 0. 5. 10. Rew. ard. PRIORADAPTTRAINbsc-trn&extd-tstextd-trn&tst. Figure 2: Different adaptation strategies. -
SYSTEM COMBINATION WITH LOG-LINEAR MODELS J. Yang, C. Zhang, ...
mi.eng.cam.ac.uk/~mjfg/yang_ICASSP16.pdf12 Jul 2016: the Gaussian sufficient statistics[19] and HMM mean and variance statistics [20]. ... 1117–1120. [20] Georg Heigold, Ralf Schlüter, and Hermann Ney, “On theequivalence of Gaussian HMM and Gaussian HMM-like hid-den conditional random fields.,” in -
CAMBRIDGE UNIVERSITY ENGINEERING DEPARTMENT DISCRIMINATIVE…
mi.eng.cam.ac.uk/~mjfg/gales_tr605.pdf13 Aug 2008: 5 Results. The performance of the proposed scheme was evaluated on the AURORA 2 task [20]. ... Two forms of front-end were examined, one HTK-based, the other ETSI-based [20]. -
THE CAMBRIDGE UNIVERSITY SPOKEN DOCUMENT RETRIEVAL SYSTEM S.E.…
mi.eng.cam.ac.uk/reports/svr-ftp/johnson_icassp99.pdf8 Mar 2000: Unstopped Term Error. Unstopped Word Error. 20 25 30 35 40 45 5020. ... 0 10 20 30 40 50 60 700.52. 0.54. 0.56. 0.58. -
Towards Learning Orientated Assessment for Non-native Learner Spoken…
mi.eng.cam.ac.uk/~mjfg/ALTA/presentations/ALTA_Sheffield_20190306.pdf21 Feb 2022: 400 hour BULATS training set. 20. AM LM % WER. -
Kate Knill ALTA Institute, Cambridge University Engineering…
mi.eng.cam.ac.uk/~kmk/presentations/Knill_ISCSLP_Keynote_2022.pdf19 Jun 2023: 20. Automatic Speech Recognition (ASR). • Hybrid ASR1. • Acoustic model:. • -
LEARNING BETWEEN DIFFERENT TEACHER AND STUDENT MODELS IN ASR ...
mi.eng.cam.ac.uk/~mjfg/ALTA/ASRU2019_TS.pdf20 Dec 2019: Furthermore, sequence-level training often yields abetter performance than frame-level training when training towardthe reference transcriptions [20]. ... The diagonal-covariance GMM for AMI-IHM had 20 mixture components perstate and used 13-dimensional -
System Combination with Log-linear Models
mi.eng.cam.ac.uk/~mjfg/icassp16_yang.pdf5 Apr 2016: the Gaussian sufficient statistics[19] and HMM mean and variance statistics [20]. ... 1117–1120. [20] Georg Heigold, Ralf Schlüter, and Hermann Ney, “On theequivalence of Gaussian HMM and Gaussian HMM-like hid-den conditional random fields.,” in -
MULTI-DOMAIN DIALOGUE SUCCESS CLASSIFIERS FOR POLICY TRAINING David…
mi.eng.cam.ac.uk/~sjy/papers/vsgm15.pdf20 Feb 2018: 1All models were built in Python using Theano [19, 20]. Stochastic gra-dient descent per dialogue was used during backpropagation to train eachmodel, and almost no optimisation of hyper-parameters (learning ... 20] Frédéric Bastien, Pascal Lamblin, -
paper.dvi
mi.eng.cam.ac.uk/~mjfg/yw293_ASRU11.pdf19 Jan 2012: µ(m)sz = g̃(µ. (m)ye , µ̃l) (20). µ(m)z = J. (m)ye µ. ... Initially, acoustic models were compensatedbased on an initial estimation of the additive noise, using thefirst and last 20 frames of each utterance. -
Face Set Classification using Maximally Probable Mutual Modes Ognjen…
mi.eng.cam.ac.uk/reports/svr-ftp/arandjelovic_ICPR06.pdf29 Apr 2006: λ(1)Dj = λ(2)Dj. (19). Then, writing. |Ci| =D. j=1. λ(i)j , (20). ... 12] K. K. Sung and T. Poggio. Example-based learning for view-basedhuman face detection.PAMI, 20(1), 1998. -
A CONFIDENCE-BASED APPROACH FOR IMPROVING KEYWORD HYPOTHESIS SCORES…
mi.eng.cam.ac.uk/~mjfg/seigel_ICASSP2013.pdf13 Jun 2013: The model alsoallows for continuous features to be represented using spline featurefunctions [20], which have been shown to yield good performancefor confidence estimation with CRFs in previous work [16]. ... 45, no. 1-3, pp. 503–528, Aug. 1989. [20] -
Practicable Assessment of Cochlear Sizeand Shape from Clinical CT ...
mi.eng.cam.ac.uk/reports/svr-ftp/gee_tr004.pdf27 Jul 2020: 9. 0 5 10 15 20 25 30 35 40. distance along outer wall (mm). ... 0 5 10 15 20 25 30 35 40 45. distance along outer wall (mm). -
A Sparse Probabilistic Learning Algorithm for Real-Time Tracking…
mi.eng.cam.ac.uk/reports/svr-ftp/williams_iccv03.pdf23 Nov 2003: Translation of magnitude lessthan 20 pixels should yield a score between 1, as Figure2 shows. ... x. y. 0 5 10 15 20-5-10-15-20. 0. 5. 10. 15. -
Recent Developments at Cambridgein Broadcast News Transcription D.Y.…
mi.eng.cam.ac.uk/research/projects/EARS/pubs/kim_rt04.pdf15 Feb 2005: Cambridge UniversityEngineering Department. RT04 EARS workshop 20. Recent Developments at Cambridge in Broadcast News Transcription. ... 0.16 0.18 0.2 0.22 0.24 0.2612. 14. 16. 18. 20. 22.
Refine your results
Date
- 350 2018
- 189 2005
- 47 2000
- 44 2006
- 28 2023
- 27 2022
- 26 2004
- 26 2003
- 25 2015
- 23 Uncertain
- 22 2007
- 20 2016
- 20 Past year
- 17 2019
- 17 2008
- 17 2020
- 15 2010
- 14 2014
- 13 2013
- 12 2001
- 11 2017
- 11 2009
- 10 2024
- 10 2021
- 8 2012
- 8 2011
- 8 2002
- 7 Past 3 months
- 7 Past 6 months
- 5 1999
- 3 Past month
- 2 Past week
- 2 Past fortnight
Search history
Recently clicked results
Recently clicked results
Your click history is empty.
Recent searches
Recent searches
Your search history is empty.