MODELLING SYLLABLE CHARACTERISTICS TO IMPROVE A LARGE VOCABULARY CONTINUOUS SPEECH RECGONISER
M. Jones and P.C. Woodland
September 1994
The acoustic-phonetic modelling used in state-of-the-art large vocabulary continuous speech recognisers (LVCSR) cannot effectively exploit the prosody based distinctions known to exist at thesyllable level. These distinctions are between the strength of the syllable (strong or weak) and the stress (stressed or unstressed) it is given.
This paper shows how a small set of syllable-sized Hidden Markov Models (HMMs) can model syllable type effectively. These models have been applied to a large vocabulary continuous speech recogniser and a 23% reduction in word error rate was achieved.
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.