摘要 |
PROBLEM TO BE SOLVED: To speed up a speech recognition processing. SOLUTION: A likelihood calculation part decides whether a power system feature speed parameterΔp(t) read out of a speech feature quantity storage part is equal to a specified thresholdαor less. When deciding that the power system feature speed parameterΔp(t) is equal to the specified thresholdαor less, the likelihood calculation part uses output probability b(i-1)(x(t-1)) or bi(x(t-1)) of a time window F(t-1) instead of output probability bi(x(t)) to calculate cumulative likelihood Lp(i, t) and cumulative likelihood Ls(i, t), writes the larger cumulative likelihood L(i, t) between the calculated cumulative likelihood Lp(i, t) and cumulative likelihood Ls(i, t) to a cumulative likelihood storage part to update the cumulative likelihood stored in the cumulative likelihood storage part. Thus, when the power system feature speed parameterΔp(t) is equal to the specified thresholdαor less, calculation of a simple normal distribution Pm(x(t)) and output probability bi(x(t)) by the likelihood calculation part is omitted, so that the speech recognition device can perform speech recognition fast. COPYRIGHT: (C)2005,JPO&NCIPI
|