Search

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

2 Mar 2000: 20.0%. 19.3%. F2. 16.1%. 3.4%. F3. 5.1%. 4.3%. F4. 4.9%. 28.2%. ... F3. F4. F5. FX. P1. tg. N/N. 19.9. 10.9. 20.5. 29.6.

1 Mar 2000: On the BNeval97 only 0.18% of speech frames were lost, which is equivalent to 20.18 seconds. ... S. 0.56. 98.24. 1.20. T. 0.00. 1.19. 98.81. a). M. S.

14 Aug 2000: The results from adding 6 new terms to each query from the 20 top documents with UBRF is given in Table 3 which shows that this is a very effective strategy.

3 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.

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.

1 Mar 2000: The Intel C compiler (version 2.4) seemed particularly efficient with the decoder running almost 20% faster than when compiled with gcc v2.7.2. ... These were trained in the same way as the first pass models but consisted of 13428 distinct models sharing

2 Nov 2000: 149.36. 1003. 50. -. 218.47. 195.72. 217.10. 192.20. 1003. 50. 158.53. ... 215.31. 261.73. 209.20. 503. 250. 158.00. 148.52. 157.44. 146.53. 1003. 250. -.

1 Mar 2000: We have found that using the full system with adaptation results in a 20% decrease in word error rate on broadcast news. ... 22.0. 21.2. F2. 18.8. 22.2. F3. 19.5. 20.5. F4. 27.3. 32.7.

1 Mar 2000: 30.5%. 20.0%. F2. 16.2%. 16.1%. F3. 6.2%. 5.1%. F4. 14.1%. 4.9%. ... F1. 21.6. 20.8. F2. 35.6. 34.9. F3. 34.1. 32.4. F4. 26.2.

12 Oct 2000: Decision tree state clustering [20] was used to select a set of context-dependent equivalence classes. ... P4b. 21.9. 33.7. 27.8. 0.287. P5a. 20.3. 32.7. 26.6. P5b. 21.0.

25 Oct 2000: The imageswere capturedusing a Fuji-700 digital camera,and about 20-30 pointcorrespondenceswereenteredmanually. The imageswereprojectively calibratedusing[3], and were then upgradedto a metric calibration using [7]. ... Notice how the

19 Apr 2000: This transcription, denoted. HTK-p2. , gave a WER of 20.5% on the scored subset. ... Post. RS. NRS. NS. #Dup. # windows. 0.1. B. 96.5. 44.20.

2 Mar 2000: 54.40. 60.00. 54.20. 63.48. 8 = 5 - GP. 54.95. 59.13.

10 Apr 2000: The results are shown in Figure 6. 0 10 20 30 40 50 60 70 80 90 100. ... 5. 5.2. 5.4. 5.6. 5.8. 6. Tra. inin. g. 0 10 20 30 40 50 60 70 80 90 100.

30 Mar 2000: 2.9. 20.5. LIMSI. 82.0. 14.6. 3.4. 3.5. 21.5. CUHTK-p1. 77.3. 18.5. ... 12.4. 18.0. 83.07. 52.32. 48.93. ----. CUHTK-s1. 20.5. 27.8. 45.89. 55.29.

30 Jan 2000: The apparentcontoursweretracked by usingcubic B-splinesnakes[4](seefig. 2). Fundamentalmatrices[13, 20] betweentwo successive imageswereformedfrom thecorrespondingprojectionmatricesandcorrespondenceswerethenfoundby solv-ing for ... IEEE

27 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.

8 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.

8 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.

10 Apr 2000: Chosen Points. 15 20 25 30 35 401.25. 1.3. 1.35. 1.4. ... Chosen Points. 20 30 40 50 60 70 80 901.45. 1.5.

8 Mar 2000: On theBNeval97 only 0.18% of speech frames were lost, which is equiva-lent to 20.18 seconds. ... M/sil S male T male S female T female. M/N 67.43 20.65 0.00 11.77 0.15S male 0.52 96.34 0.47 2.51

10 May 2000: Table 3: Effects of Query Expansion on MAP. The results from adding 6 new terms to each queryfrom the 20 top documents with UBRF is given in Table 3which shows that

19 Apr 2000: How-ever the transcription systems used (see e.g. [6]) typicallyrun rather slower than real-time and have a word errorrate of over 20%. ... Method CPUs non-news newswindow, 5@1s 3514 5459s=60.8% 110s=0.5%window, 10@2s 967 5480s=61.0% 114s=0.5%window,

5 May 2000: 0.1. 0.5. 2. 5. 10. 20. 3040506070. 80. 90. 95. 98. ... 99.5. 99.9. 0.1 0.5 2 5 10 20 30 40 50 60 70 80 90 95 98 99.5 99.9.

8 Mar 2000: The Intel Ccompiler (version 2.4) seemed particularly efficient with thedecoder running almost 20% faster than when compiled withgcc v2.7.2. ... 22. 23. 24. 25. 26. 27. 28. 29. 30. 0 5 10 15 20 25.

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.

8 Mar 2000: Thelargest gains come from the most severely mismatched condi-tions: NM/RB improves by 30% and BM/FB by 20%, whilethe NM/FB data improves by just 7%. ... It was noted that these perplexities, although about 20%higher than typically observed on read

1 Jun 2000: find, and ;. compute the distance between and ;. update to minimize the function in (20);end while. ... 34. 26. 33. 27. 32. 28. 31. 2930. 5 10 15 20 25 30 350.

10 Apr 2000: that using the full system with adaptation resultsin a 20% decrease in word error rate on broadcast news. ... Data SegmentationType PE Segments S1 SegmentsF0 11.1 11.9F1 22.0 21.2F2 18.8 22.2F3 19.5 20.5F4 27.3 32.7F5 29.5 31.0FX

8 Mar 2000: Focus Proportion of dataCond BNdev96ue BNeval97. F0 22.3% 45.0%F1 30.5% 20.0%F2 16.2% 16.1%F3 6.2% 5.1%F4 14.1% ... F0 13.3 13.0F1 21.6 20.8F2 35.6 34.9F3 34.1 32.4F4 26.2 25.7F5 29.0 27.5FX 50.9 46.8.

10 Apr 2000: In [20] a reconstruction method based on parameterising the surface by radial curveswas developed. ... deg. 5 10 15 20 25 30 359. 9.2. 9.4. 9.6.

5 Oct 2000: 27.8 0.287P5a 20.3 32.7 26.6P5b 21.0 32.8 26.9 0.292P6a 20.3 32.6 26.5 0.284. ... Speech Transcription Workshop, College Park. 20. S.J. Young, J.J. Odell & P.C.

6 Sep 2000: Stage2 For eachGaussianÒ we test a number(20 in this case)of randomlychosenGaussians,asin Stage1. ... Table1 andTable2 show the resultsof FD on the singleand6 Gaussianper statesystems.ThesingleGaussiansystemshows anoverall decreasein WER of 20.3%after4

17 Oct 2000: 2 tg 21.06 9.60. Lattice Gen. bg 22.12 10.88Lattice Gen. tg 17.20 7.88.

8 Mar 2000: ÂJÙYÉÀÃEùZ@-MÐZÃ;ÄÂcÏcÃ79ÄzÃ Ø À¿ÅÀÌZÆÌZÆÇzkÃ3Þß.à.Ý%î#ïEÞWÇzà ¿TÇ3Ø ÌcÉÀÉ=Ã;Ø;ÇzÀÌZÆNÚoÊÇzÍà ÂcÆNÆÈkÍ«Ù/Ã;Ä7ÌcËÒÈYÃ;ÄÅÓÇzÃÑÄzÍ¿=ÆNÁ|ÂJÄzÃ;ÆÇzkà ¿ÅÀ¿. 0 10 20 30

10 Apr 2000: This transcription, denotedHTK-p2 , gave a WER of 20.5% on the scoredsubset. ... which usethis strategy and have a lower word error rate of 20.5%.

10 Apr 2000: Rec. HTK cc-proc HTK-p1 LIMSI NIST-B2Error 55.51 37.93 67.86 57.20 62.46Rec. ... 6Another method of modifying the TER to model retrieval weighting moreclosely can be found in [20].

10 Apr 2000: 60.00 52.56 59.137 = 5 - WP 54.40 60.00 54.20 63.488 = 5 - GP 54.95 59.13 53.56 60.87.

8 Mar 2000: Unit Gradient Reference. Stopped. Stopped/Stemmed. 20 25 30 35 40 45 5015. ... Refe. renc. e. 0 20 40 60 80 100 1200.36. 0.38.

26 Apr 2000: the skeleton representation [2]and union of spheres representation [20]), or to volume data [24]. ... 00:01:00 00:30:00 00:00:30Frames 20 20 20 20Polygons per frame 4, 000 38, 000 33, 000 3, 300Morph creation time 00:01:31 00:45:20

12 May 2000: The actual orientation of these sweeps varied somewhat between examinations.Each complete examination was performed in a single breath hold (roughly 20 seconds). ... B-scans plane positionSegment 20. B-scans. Figure 5: Volume measurement from multiple

31 Jul 2000: Trees based onTIMIT alone are much less eective on theSwitchboard test set (20%), but adding them to theSwitchboard training data (ICSITIMIT) resultsin a small additional gain (32%). ... 20%). 0.34! 0.17 (51%). ICSI 0.72! 0.50. (30%). ICSITIMIT 0.71! 0.48

6 Nov 2000: In this caseexamples can be constructed where alternativediscrimina-tive training schemes such as the Maximum Mutual Infor-mation Estimation (MMIE) can provide better performancethan MLE [20]. ... 20–23, New York. [9] P.S. Gopalakrishnan, D. Kanevsky, A

10 Apr 2000: 4 Analysis with xpert 15. 5 Conclusions 20. 6 Acknowledgements 20. ... 20 Acknowledgements. 5 CONCLUSIONS. xpert has been shown to allow the user to quickly identify the location of all the errors in a transcrip-tion.

10 Apr 2000: 44.32Condition Match(%) 49.75 18.72 20.20 15.27 35.96 39.90 37.93 54.68 44.83Total Match(%) 52.50 7.50 15.62 10.62 ... 1. 3. 4. 5. 6. 4 8 120 16 20. Ward 2D (on mean).

31 Jul 2000: SeeGodfrey, Holliman andMcDaniel (1992) for a detailed corpus description.] A vocabulary of approximately 20 000. ... set). Transcription type Models Phone error rate Deviation from baseformsDictionary baseforms — 33.6% 0%Automatic [Step (6)] Standard