Search

6 Oct 2008: each test set– Whole-word models, 16 emitting-states with 3 components per state. • ... System Test SetA B C. VTS 9.84 9.11 9.53+ SVM 7.52 7.35 8.11.

16 Jun 2008: During recognition the following procedure is used:1. Compensate the acoustic models for the test noise con-. ... dition2. Recognise the test utterance Y to obtain 1-best hypoth-. esis, h = h1,. ,

13 Aug 2008: In addition tobackground additive noise convolutional distortion was added to test set C. ... For all SNRsand test sets gains over the optimised VTS scheme were obtained.

6 Nov 2008: In addition tobackground additive noise convolutional distortion was added to test set C. ... For all SNRsand test sets gains over the optimised VTS scheme were obtained.

11 Jan 2008: Two test sets defined by BBN were used for evaluating the sys-tems. ... It was ensured that therewas no overlap between this training and test data.

16 Feb 2008: 120,000 instances in the training set. 8,000 instances in the test set. ... Lattice-based MMIE was performed. Test:. Test set is 8400 utterances ( 25 hours) from 10 heldout speakers.

11 Jan 2008: 3.3. Test datasets. The test sets used to evaluate the systems are bnmdev06 and bcmdev05.bnmdev06 comprises 3.6 hours of data taken from a range of BNsources. ... It includes some of the standard existing test sets describedin [10], dev04f, eval03m and

6 May 2008: NIST 2008 Arabic-English MT evaluation development 8. Lowercase BLEU scores over three test sets from 2002 through 2006:. ... 8 nist.gov/speech/tests/mt/2006 nist.gov/speech/tests/mt/2008Department of EngineeringUniversity of Cambridge. NIST Open Machine

13 Jun 2008: If something is known about the possible test acoustic environment– multi-style (multi-environment) training may be used– “clean” model trained under a variety of conditions– also helps general robustness. • ... 20dB used for SPR

3 Dec 2008: The ML-SAT test procedure is first run toobtain initial ML speaker transforms. ... There is no need to estimate speaker-specific discriminative transforms on the test data.

11 Jan 2008: However for many situations there may be limited,or even no, test adaptation data available. ... For the eval03 test set the average utterance lengthwas 3.13 seconds, compared to the average side length of 153.75seconds.

26 Mar 2008: The 2006 are theevaluation results on the entire “NIST” portion of the 2006 test set. ... On these test sets with. 5The 2006 evaluation results are based on mixed case output, the devel-opment results are lower-case.

16 Jun 2008: Resolv-ing the mismatch between the training and test acoustic con-ditions has been an active area of research for many years. ... All results are averaged over threeof the four available test sets, Feb89, Oct89, and Feb91, a totalof 30 test speakers and

26 Mar 2008: Two test sets defined by BBN were used for evaluating the sys-tems. ... It was ensured that therewas no overlap between this training and test data.

16 Feb 2008: Test Sets: SWB1 (1831 utterances) and SWB2 (1755 utterances). MBR decoding strategy: A search on lattices. ... Test Set: Chinese-English NIST MT Task (2002) , 878 sentences, 1000-best lists.

11 Jan 2008: Results are given on three test sets: bnat05 (5.72 hours)and bnat06 (2.76 hours) are broadcast news data, with the for-mer being more closely matched to the training ... When the combination of S0 and D1 is performed, gainsare seen for all test sets.

27 Aug 2008: Horse detection. ROC curve - 3.4% EER. 100 training images. 200 test images.

11 Jan 2008: This paper investigates using the MBR criterion for unsuper-vised discriminative language model adaptation in test-time for speechrecognition and translation systems. ... Unsupervised test-time discriminative adaptation of mixture languagemodels was

17 Jan 2008: At each test point, awindow is defined with the point at its centre. ... For this particular data set, the 5% test eliminates all but three of the original twentyseeds.

27 Aug 2008: Horse detection. ROC curve - 3.4% EER. 100 training images. 200 test images.