The role of speech recognition of address coding

摘要

In mail processing, address data is either automatically extracted via OCR and address interpretation or, in case of rejects, entered manually by video coding operators. Currently, the necessary data is keyed in via keyboard. Speech recognition is an alternate way of entering data because adequate hardware resources are available and recognition algorithms have become more robust. The advantages of speech are its natural mode and its easy operation.rnThe essential requirement for speech recognition is recognition accuracy because the higher the recognition rates, the more mail can be processed with speech input and, hence, the more time can be saved. Dedicated systems currently attain a recognition accuracy of some 80% on a lexicon size of 1000 entries, a result which deteriorates to 70% in case of 10000 entries (German, speaker-independent). Off-the-shelf products, like ViaVoice or Dragon Dictate, are not suitable for the task of address coding as their recognition rate is too low for this highly specialized application.