SPEAKER DEPENDENT KEYWORD SPOTTING FOR ACCESSING STORED SPEECH
Kate Knill and Steve Young
October 1994
This report investigates the use of a speaker-dependent HMM word-spotter to retrieve spoken messages. The baseline word-spotter consists of a parallel network of keyword and background filler models. A further pass, using the filler models only, can be used to re-score the putative keyword hits. Word-spotting performance using (i) whole-word and (ii) sub-word keyword models is investigated. For each keyword model type, effects on performance of re-scoring, Gaussian component number, and parameter set are evaluated on a set of spoken messages, taken from the Video Mail Retrieval database. Overall, sub-word models are shown to yield a higher hit rate, particularly before the first false alarm occurs.
If you have difficulty viewing files that end '.gz'
,
which are gzip compressed, then you may be able to find
tools to uncompress them at the gzip
web site.
If you have difficulty viewing files that are in PostScript, (ending
'.ps'
or '.ps.gz'
), then you may be able to
find tools to view them at
the gsview
web site.
We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.