S.-Y. Lung (Taiwan)
Speaker identification, Karhunen-Loevetransform, Adaptive threshold.
When using thresholding method to eigenspace derived from the Karhunen-Loeve transform, a fixed threshold is not suitable if the intra-speaker and inter-speaker data variation is large. Here, we propose a new variation thresholding method. The method requires two thresholds to be automatically selected. Furthermore, the variation thresholding can be use to any speaker data derived from Karhunen-Loeve transform. It is demonstrated that 93% correct classification rates can be achieved by the use of the first 32 variation features.
Important Links:
Go Back
