Information Retrieval from Spoken Documents

This paper describes a designed and implemented system for efficient storage, indexing and search in collections of spoken documents that takes advantage of automatic speech recognition. As the quality of current speech recognizers is not sufficient for a great deal of applications, it is necessary to index the ambiguous output of the recognition, i. e. the acyclic graphs of word hypotheses — recognition lattices. Then, it is not possible to directly apply the standard methods known from text-based systems. The paper discusses an optimized indexing system for efficient search in the complex and large data structure that has been developed by our group. The search engine works as a server. The meeting browser JFerret, developed withing the European AMI project, is used as a client to browse search results.

[1]  W. Russell,et al.  Continuous hidden Markov modeling for speaker-independent word spotting , 1989, International Conference on Acoustics, Speech, and Signal Processing,.

[2]  Lukás Burget,et al.  Phoneme Based Acoustics Keyword Spotting in Informal Continuous Speech , 2005, TSD.

[3]  Lukás Burget,et al.  Comparison of keyword spotting approaches for informal continuous speech , 2005, INTERSPEECH.

[4]  Roeland Ordelman,et al.  Transcription of conference room meetings: an investigation , 2005, INTERSPEECH.

[5]  Andreas Stolcke,et al.  The ICSI Meeting Corpus , 2003, 2003 IEEE International Conference on Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03)..

[6]  Sergey Brin,et al.  The Anatomy of a Large-Scale Hypertextual Web Search Engine , 1998, Comput. Networks.

[7]  Walter Daelemans,et al.  Data-Oriented Methods for Grapheme-to-Phoneme Conversion , 1993, EACL.