摘要 |
<P>PROBLEM TO BE SOLVED: To solve a problem in which a quality estimation error becomes large owing to differences in conversation style among a "lecture", a "meeting", a "free conversation", and "data verification". <P>SOLUTION: A feature extraction part 21 analyzes a conversation voice signal to extract the frequency Rn of word exchange in the conversation as feature information and a decision part 22 decides a lecture from Rn<α, a meeting from α<Rn<β, a free conversation from β<Rn<γ, or data collation from γ<Rn; and individual quality estimation expressions by quality factors determined by conversation styles are selected based upon the decided conversation style and calculated by an estimation part 24 as to an inputted corresponding quality parameter to find individual quality estimated values by the factors, and total quality estimation expressions determined by the conversation styles are selected from a database part 25 based upon the decided conversation style and calculated as to the estimated individual quality estimated values to output a total quality estimated value. <P>COPYRIGHT: (C)2005,JPO&NCIPI |