"... In PAGE 21: ... Our best results came from this nal run using fewer features. These results appear in Table5 . As before, the results on WF2 were produced by training on the entire WF1 data set, and then testing on WF2.... ..."

"... In PAGE 6: ... Possible solutions are to em- ploy IDDQ testing [Ack83], apply more accurate circuit simulation of faults, detect the bridge as a delay fault, or redesign the cells so that a discrepancy is guaran- teed for at least one input combination for each cell. In Table5 we show the results of our system generating IDDQ test patterns for same set of bridging faults that produced the results in Table 4. We also plan on integrating the testing for breaks on the interconnection lines, and the testing for defects within the cells to Carafe and Nemesis.... ..."

"... In PAGE 11: ... We estimated the reliability of the predictions (and the transition matrix model in general) by con- structing the future states proceeding from differ- ent transition matrices and initial states of the same study area. The Keava study area was taken as an example and three different forecasts were made based on: I) the matrix of the 31 year period, KN5 182 and KA5 182; 11) the matrix of the short (16 year) period, KN6682 and KA6682, and 111) the averaged matrix, KN5166 + KN6682 and KA5166 + KA6682 (see Table7 ). As we can see, the dis- crepancies between different predictions are on the level of 1% (with respect to the total study area), whereas the changes themselves reach 1.... ..."

"... In PAGE 10: ...2 More powerful tests Despite its simplicity and intuitive appeal, the TOST su ers from a lack of power. The line labeled TOST in the top part of Table1 shows the power function, P(reject H0), for parameter points with T ? R = U (or L), points on the boundary between H0 and Ha. The power function is near for 2 D near 0, but decreases as 2 D grows.... In PAGE 10: ... The TOST is clearly biased. The bottom part of Table1 shows the power function when the two drugs are exactly equal, T = R. The power is near one for 2 D near zero, but decreases to zero as 2 D increases.... In PAGE 13: ... In this description, we assume 1=2 gt; gt; 1 ? F(3 =4). Brown, Hwang and Munk (1995) in their Table1 show that if r 4, then = :05 gt; . The new test for is given in the Appendix.... In PAGE 15: ... But R is slightly biased whereas the Brown, Hwang and Munk test is unbiased. A small power comparison of the TOST, Brown, Hwang and Munk test, and our new test is given in Table1 for = :05 and r = 30. In the top block of numbers, T ? R = .... In PAGE 17: ...8 when T = R. In this case, Table1 indicates there is no advantage to using the new tests over the TOST. But if the variability turns out to be larger than expected in the planning stage, the new tests o er signi cant power improvements.... ..."

"... In PAGE 1: ... Data base, features and model The data base used in these experiments was the same as in the previous paper [1] extended with vocal music and choir music. Table1 . Sound database overview.... ..."