5th International Conference on Spoken Language Processing
Directory assistance systems are amongst the most challenging applications of speech recognition. Today, complete automation of the service fails because of the lacking accuracy of current speech recognizers, which are simply not able to differentiate between hundreds of thousands or even millions of different names occurring in large cities. In this paper, we show that this situation can be remedied by systematically combining all available knowledge sources (last names, first names, street names, partly including their spelled versions) in a statistically optimal way. Especially designed confidence measures for N-best lists are proposed to detect misrecognized turns. Applying these techniques in a hierarchical setup is judged as the enabling step for automating large scale directory assistance. In first experiments, we e.g. are able to service 72% of the inquiries for a database of 1.3 million entries with a remaining error rate of only 6% (or 62% with an error rate of 2%).
Bibliographic reference. Kellner, Andreas / Rueber, Bernhard / Schramm, Hauke (1998): "Using combined decisions and confidence measures for name recognition in automatic directory assistance systems", In ICSLP-1998, paper 0454.