Speech:Exps 0304 030

Description
Author: Stephen Thibault (UserID: sdt1001)

Date: 3-14-2018

Purpose: 300hr Train Data with sphinx_train.cfg from 0301/006 without LDA and with 8000 senone count for 300hr.

Details: Language Model not created here, but in the Decode and Scoring so that this can be used with dev.trans and eval.trans. Performed on drone server Obelix. Unseen Decode and Scoring performed in 0304/031.

Results: In progress.

UPDATE: ABANDONED. Determined on 17 April that just activating LDA in the sphinx_train.cfg file is NOT the only necessity to train with LDA correctly. So the results of 28.4% WER obtained by 2017 Spring Rebels Team in 0301/011 are irrelevant.