"Factors that are important or might be important to the success of MLP based features and updates on current efforts"
While we achieved good results by using MLP-based feature in LVCSR task, we want to understand why it is good and how to make it better, both theoretical and practical. This involves not only analysis but also a lot of tuning. Some are important, some are not so effective. I will update you on my involvement in the NA project for the past several months. Most of the slides come from the NA retreat '04. There will be some repetition and extension from the talk I gave two weeks ago on the ICSLP poster.