Speech Enhancement with Factor-Graphs
Speech Enhancement with Factor-Graphs
The human auditory system has the ability to analyse and separate complex sounds into single sound sources. Due to this ability we are able to understand speech in noisy environment. People who have hearing defects can't do so.
Probability models with factor graphs are a new possibility to develop algorithms for speech processing. In this diploma thesis I try to show how algorithms with factor graphs can be used concerning these problems. An existing model, a simulation of the LPC-model, should be extended and optimised for real assumptions of speech signals. One possibility is to implement a TVAR-model for filter coefficients and the fundamental frequency, so that this two components aren't constant any longer.