DREU SUMMER 2012

Journal Week 9 Now that we have clean data, I began working on extracting features from the speech. I tweaked an existing praat script to extract intensity - a major vocal attribute that has to do with speech volume. I ran the script on the cleaned up session (using the text grids that I generated last week), and then on all 76 sessions, in order to do some preliminary statistical analysis. We then calculated the intensity at the beginnings and ends of turns in conversations, using speaker identification tags (done by the OYEZ project) to determine when turns begin and end. This is helpful in measuring local entrainment - we will calculate the differences in intensity measures to see at which points people are entraining. I also loaded another Supreme Court session to AMT, and all 38 HITs were done within an hour! I incorporated gold standard questions to check that the HITs are being done correctly. This week the lab went on another trip, this time to The High Line and Chelsea Market! Here is a picture of me and Mandy, the other DREU student in the lab: People are beginning to give their final presentations at the lab. This week Mandy spoke about her work this summer on Wordseye with Twitter - using tweets to generate 3D scenes. Alex, another student, spoke about using targeted clarification questions to improve errors in ASR systems. It's hard to believe that the summer is coming to an end! [PREVIOUS] [NEXT]