| US 7,539,616 B2 | ||
| Speaker authentication using adapted background models | ||
| Zhengyou Zhang, Bellevue, Wash. (US); and Ming Liu, Urbana, Ill. (US) | ||
| Assigned to Microsoft Corporation, Redmond, Wash. (US) | ||
| Filed on Feb. 20, 2006, as Appl. No. 11/358,302. | ||
| Prior Publication US 2007/0198257 A1, Aug. 23, 2007 | ||
| Int. Cl. G10L 15/00 (2006.01) | ||
| U.S. Cl. 704—233 [704/239; 704/255; 704/256] | 12 Claims |
| 1. A method comprising:
receiving a speech signal produced by a user;
forming adapted means for each of a plurality of mixture components by adapting a background model comprising background means
for each of the plurality of mixture components based on the received speech signal;
receiving a nominal user identification;
determining a similarity score between the received speech signal and a training speech signal provided by the nominal user
by determining the sum of functions determined for the plurality of mixture components, wherein each function comprises the
product of a posterior probability of a mixture component based on the received speech signal and a difference between an
adapted mean and a background mean; and
using the similarity score to determine if the received speech signal was produced by the nominal user.
|