| 摘要 |
For processing recognized speech a method is proposed, which comprises the steps of receiving and/or generating (S 0 ) recognized speech data (RSD), determining and/or generating (S 1 ) a current measure of confidence (CMOC) for at least possibly erroneously recognized words of or within said recognized speech data (RSD), wherein an iterative process is performed comprising the following steps to be performed consecutively until a stopping condition (S 3 ) is met: indicating/marking (S 2 , S 5 -B) at least some of said possibly erroneously recognized words as current indicated/marked errors (CIME) using said current measure of confidence (CMOC), receiving (S 4 ) selection/correction data (SCD) being descriptive for a selection/correction with respect to said current indicated/marked errors (CIME), determining (S 5 ) and/or generating an improved measure of confidence (IMOC) at least for further or remaining possibly erroneously recognized words of or within said recognized speech data (RSD). If the stopping condition is met, an improved recognition result (IRR) is output (S 8 ).
|
