Speech:Exps 0305 034

Description
Author: Rose Salemi (UserID: ras1002)

Date: 4-17-2018

Purpose: This experiment combines two elements: using LDA and using three alternate scripts in place of parseLMTrans.pl to remove bracketed words from the transcript, dictionary and LM, with the exception of the three filler words [NOISE], [VOCALIZED-NOISE] and [LAUGHTER].

Used with seen data, AKA test-on-train data. 300h hour corpus

Details: Steps for this are on my log page https://foss.unh.edu/projects/index.php/Speech:Spring_2018_Rosali_Salemi_Log#Week_Ending_April_16.2C_2013

Results:

ON HOLD until other experiments have been run that will determine the best seed value to put in the MLLT python script in the random function referred to in my log on 4/13/18 and 4/14/18. If we don't get to it before the end of the semester, follow the steps in the link above on 4/16/18.