Table 1: Sentences used in the test set
Next: Automatic scaling of
Up: THE BASIC SYSTEM
Previous: THE BASIC SYSTEM
Table 1: Sentences used in the test set
At the time of writing, only the prototype TIMIT database was available. This contains all the training material of the full database but none of the test material. Thus, it was necessary to partition this database into training an testing portions. There are 420 speakers in total which were divided into 317 speakers for training and 103 speakers for testing. Eight sentences were used per speaker (the si and sx sentences). The identity of the test speakers are given in table 1, those marked with an asterisk are believed to have been used by Lee and Hon for testing their Hidden Markov model recogniser [2]. The authors are very grateful to Vassilios Digalakis and Mari Ostendorf of Boston University for their help in compiling this list.