摘要 |
A method is provided for deriving recognition reference data for use in pattern recognition and, in particular, speech recognition. According to the invention, recognition reference data are provided comprising a first part representing valid items of speech or non-speech sounds, and a second part derived from those portions of an input signal known to give rise to recognition errors. In particular, reference data are provided to model insertion errors and substitution errors known to occur in recognition results.
|
申请人 |
BRITISH TELECOMMUNICATIONS PUBLIC LIMITED COMPANY;RINGLAND, SIMON, PATRICK, ALEXANDER;TALINTYRE, JOHN, EDWARD |
发明人 |
RINGLAND, SIMON, PATRICK, ALEXANDER;TALINTYRE, JOHN, EDWARD |