This thesis contains new FPGA implementations of adaptive signal segmentation and autoregressive modeling techniques. Both designs use Simulink-to-FPGA methodology and have been successfully implemented onto Xilinx Virtex II Pro device. The implementation of adaptive signal segmentation is based on the conventional RLSL algorithm using double-precision floating point arithmetic for internal computation and is programmable for users providing data length and order selection functions. The implemented RLSL design provides very good performance of obtaining accurate conversion factor values with a mean correlation of 99.93% and accurate boundary positions for both synthesized and biomedical signals. The implementation of autoregressive (AR) modeling is based on the Burg-lattice algorithm using fixed point arithmetic. The implemented Burg design with order of 3 provides good performance of calculating AR coefficients of input biomedical signals.
Unified Autoregressive Modeling for Joint End-to-End Multi-Talker Overlapped Speech Recognition and Speaker Attribute Estimation
