| US 7,478,043 B1 | ||
| Estimation of speech spectral parameters in the presence of noise | ||
| Robert David Preuss, Sagamore Beach, Mass. (US) | ||
| Assigned to Verizon Corporate Services Group, Inc., New York, N.Y. (US); and BBN Technologies Corp., Cambridge, Mass. (US) | ||
| Filed on Jun. 05, 2003, as Appl. No. 10/454,962. | ||
| Claims priority of provisional application 60/455748, filed on Mar. 19, 2003. | ||
| Claims priority of provisional application 60/386008, filed on Jun. 05, 2002. | ||
| Int. Cl. G10L 15/20 (2006.01); G10L 15/00 (2006.01) | ||
| U.S. Cl. 704—233 [704/231] | 42 Claims |
| 1. A speech analysis device comprising:
a plurality of sensors for receiving a signal segment;
a codebook configured to store parameters corresponding to a plurality of hypothesized speech power spectral density functions;
and
a discriminant function calculation component configured to evaluate a likelihood of at least one of the hypothesized speech
power spectral density functions corresponding to the signal segment based on application of a discriminant function evaluated
using a preconditioned conjugate gradient (PCG) process configured for block Toeplitz type matrices with a block size corresponding
to the number of sensors.
|