Recent years have seen impressive improvements in systems for automatically recognizing speech. Research in comparing the quality of different speech recognition systems often uses as a benchmark test a series of isolated words, checking how often each system makes errors recognizing the word. Table 7.25 shows an example17 of one such test, comparing two speech recognition systems, called generalized minimal distortion segmentation

(GMDS) and continuous density hidden Markov model

(CDHMM).

(a) Estimate the population proportion correct for each system.

(b) Show all steps of McNemar’s test to compare the population proportions. Interpret.

(c) Construct a 95% confidence interval to compare the population proportions. Interpret.

TABLE 7.25 CDHMM GMDS Correct Incorrect Total Correct 1921 58 1979 Incorrect 16 5 21 Total 1937 63 2000