发明名称 METHOD TO EXTRACT FEATURED VALUES FROM VOICE SIGNALS
摘要 PROBLEM TO BE SOLVED: To conduct real time processes by reliably determing formant frequencies or featured values indicating these formant frequencies from voice signals using a smaller number of computations. SOLUTION: The formant frequencies are determined by forming a power density spectrum by discrete frequencies. Then, autocorrelation coefficients are formed from the spectrum against continuous frequency segments. Then, the sum of the error values formed by the above processes is made minimum against all segments and the optimum boundary frequency of the segment is determined against the minimum value. Then, the formant frequencies are computed from the predicted value coefficients computed from the autocorrelation coefficients. Then, the minimum value of the sum of the error values against an individual segment is formed from a dynamic programming. In the programming, an auxiliary value is computed from the spectrum, stored as a table and the autocorrelation coefficients are determined from the table against individual frequency segment requried for the computations of the dynamic programming.
申请公布号 JPH1069296(A) 申请公布日期 1998.03.10
申请号 JP19970104632 申请日期 1997.04.22
申请人 PHILIPS ELECTRON NV 发明人 WELLING LUTZ;NEY HERMANN PROF DR
分类号 G10L11/00;G10L13/00;G10L15/00;G10L19/06;(IPC1-7):G10L9/02;G10L9/14 主分类号 G10L11/00
代理机构 代理人
主权项
地址