With little labor, it detects the pronunciation fluctuation case robustly, it designates that the pronunciation fluctuation rule whose pan- metaplasia is high is acquired as topic. It can solve this topic, with the pronunciation fluctuation systematic extracting equipment which has with the voice data memory section and the normal mode pronunciation memory section and the sub word language model formation department and the speech recognition section and the finite difference extraction section. The voice data memory section remembers the voice data. The normal mode pronunciation memory section remembers the normal mode pronunciation data which displays the normal mode pronunciation of the voice data. The sub word language model formation department forms the sub word language model from the normal mode pronunciation data. The speech recognition section recognizes the voice data making use of the sub word language model. The finite difference extraction section comparing with the recognition result and the normal mode pronunciation data which the speech recognition section outputs, extracts these finite difference.
展开▼