ALPHADIGITS COPRUS

• Data is distributed by the Oregon Graduate Institute (OGI)

• Consists of 75 hours of speech from 3031 speakers collected over telephone lines

• The strings consisted of six words on an average

• We extracted 12 MFCC features for all the phones from 200 frames of data using the ISIP speech recognition front end

• Ten percent of this data was used as the test set - this division of the data was done randomly

• We used the first and eighth MFCC coefficients for the 'f' and 'uw' phones for the two-class two-dimensional classification