Effects of Manual Phonetic Transcriptions on Recognition Accuracy of Streetnames

In automotive systems, the accurate recognition of street- and place names as needed e.g. for speech controlled navigation systems is difficult, since many of them do not follow regular transcription rules. We report on experiments investigating the effects of correct manual transcription versus strictly auto-matic transcription, with regard to recognition accuracy. Our experiments showed that in case of large recognizer lexica, the recognition rate could be increased by about 10% by using a manually corrected versions of the pho-netic transcriptions used for the recognition process.