*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
Title: An Indirect Speech Enhancement Framework Through Intermediate Noisy Speech Targets
Committee:
Dr. Lee, Advisor
Dr. Anderson, Chair
Dr. Moore
Abstract:
The objective of the proposed research is to develop an indirect approach to single-channel speech enhancement by converting difficult noisy speech to an intermediate representation and subsequent refinement. The research is motivated by observations that a considerable performance gap exists among enhancement results of speech in different types of background noise. In order to improve the performance of speech in challenging noise conditions, we seek to replace the background with noise types that are easier to handle. The indirect approach to speech enhancement requires identification of an appropriate conversion target, a conversion technique, and a refinement procedure. In order to identify the conversion target, we discussed the properties of noise signals that interact with a DNN-based enhancement system. We developed some qualitative guidelines of selecting intermediate targets from empirical knowledge. Three conversion techniques were compared and each one had its own usage depending on the availability of training data. Lastly, we explored potential network adaptation techniques that could be applied to the converted speech.
