SPEAKER ADAPTATION BASED ON PRE-CLUSTERING TRAINING SPEAKERS
Yuqing Gao, Mukund Padmanabhan, et al.
INTERSPEECH - Eurospeech 1997
The spectral analysis of short data segments has traditionally been done using eigenvalue-based matrix-analysis methods. Recently, some IIR adaptive filters have been used for the spectral analysis of multisinusoidal signals corrupted by noise, but these have only been used for analyzing long data segments, since they normally require the analysis of many data samples before they converge. They do have the advantages of being easy to program, do not require much memory for storage, and sometimes have few divisions. In addition, they often have very good resolution, especially for characterizing sinusoids at frequencies much less than the sampling frequency. A new IIR adaptive resonator-in-a-loop filter bank is described that can be used for high-resolution spectral analysis of not only long data segments, but short data segments as well, with accuracies approaching the Cramer-Rao lower bounds for SNR’s as small as 10 dB. The basic approach taken is to reanalyze the data segment many times while running the data forwards and backwards through the filter, as the coefficients converge. Special care is taken at the data endpoints, when reinitializing the filter state variables, to eliminate transients. © 1993 IEEE
Yuqing Gao, Mukund Padmanabhan, et al.
INTERSPEECH - Eurospeech 1997
Geoffrey Zweig, Mukund Padmanabhan
INTERSPEECH - Eurospeech 1999
Mukund Padmanabhan, Ken Martin
IEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing
Mukund Padmanabhan, George Saon, et al.
IEEE Transactions on Speech and Audio Processing