|
|
haiyan7988(金币+1, 翻译EPI+1): 2011-01-11 21:36:40
Structurally, use tandem structure to further judge speaker's voice .
According to the need, intercept speech sections with various lengths, and avoid intercepting speech section that is the same part speech data.
Statistically analyze the probability of the real speaker in the first N candidated position,take the probability as an important reference to set the value of N,and finally designate appropriate N value . |
|
回复此楼
+2/222
