Speech:Exps 0304 023

Description
Author: Stephen Thibault (UserID: sdt1001)

Date: 3-11-2018

Purpose: Replication of 0301/011 Unseen Decode and Scoring.

Details: Uses the copied Train Data from 0301/006 which resides in 0304/017 and the default Language Model created in this sub-experiment directory as well as the sphinx_train.cfg parameters from the 0301/011 Decode and Scoring including LDA. Run on drone server Asterix.

ADDITIONALLY: This was done in 0304/019, but trouble shooting the CORE DUMP described there determined that the "wav" directory was empty (and has been in every Unseen Decode performed this semester). The intent here is to manually copy over the "wav" directory from 0304/017 and see how the Decode and Scoring goes.

Performed manual copy of "wav" directory contents from 0304/017 with the following command: "cp -r wav/* /mnt/main/Exp/0304/023/wav" AFTER running "genFeats.pl -d".

Results: FAILED. Ran "sclite..." resulted in the same "Error: Not enough Reference files loaded

Missing: " then a LOT of .sph files.

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.