A system for content based retrieval of speech recordings accepts vague queries and it performs a best-match search to find speech recordings that are likely to be relevant to the queries. Content based speech retrieval is based on metadata that contains clues about the content of the speech recordings. This metadata is generated by an automatic indexing process. To achieve reasonable response times of less than one or two seconds the metadata (i.e. the set of document descriptions) has to be organized in suitable data structures.
KeywordsHide Markov Model Speech Recognition Recognition Error Speech Recognition System Indexing Feature
Unable to display preview. Download preview PDF.
- 1.The SAMPA alphabet evolved from the European SAM project which develops methodologies, tools, and databases for the assessment of multilingual speech recognition systems.Google Scholar
- 2.In phonology, a phoneme is a minimal unit differentiating between meanings. In the speech recognition literature, however, the term phoneme is used as in this chapter.Google Scholar
- 3.The Croats took a vote on independence.Google Scholar