Automatic recognition of spoken spelled names using speaker‐independent templates

An automatic word recognizer is used to accept strings of spelled letters spoken in isolation. The output of the recognizer is a set of best candidate letters for each letter spoken in the string. Candidate strings forming spelled last names and initials are constructed from the candidate letters and compared and matched to name strings stored in a telephone directory. An evaluation was carried out using speaker independent reference templates obtained by the clustering procedures described by Rabiner et al., which operated on speech samples collected from a 100 speaker data base. Ten test speakers spelled a list of 50 names extracted from the directory. The median error rate for spelled letter is approximately 30% while the median error rate for correct matching directory strings is 2%. These results are compared with earlier results using speaker dependent reference templates.