As it is well known, the successful deployment of large scale speaker independent word recognition systems for use over the telephone network can generate great cost savings, besides new revenue sources. Telefonica I+D is very interested in the development of new services that use vocal access and response. This is the reason why we have been involved in an extensive study of different structures of isolated word recognizers working with 'clean' and 'telephone quality speech'. To carry out these studies we have collected two databases: a clean database ,using a microphone and a telephone database; both databases consist of the same utterances and speakers, but recorded by two different methods. The results we are going to point out have been extracted in several tests using these two collections of files and some 'quality modified'speech files. Keywords: Hidden Markov Models, 'telephone-quality database','clean database'.
Bibliographic reference. Villarrubia, L. / Poza, M. J. / Crespo, C. (1991): "Influence of the telephone line on automatic speech recognition", In EUROSPEECH-1991, 1363-1366.