Speech:Exps 0288 001

Description
Authors: James Schumacher

Date: 3/30/16

Purpose: In our first experiment we ran a 300 hour train with CFG VARNORM set to no, FINAL NUM DENSITIES set to 32, CFG TIED STATES set to 8000 and CFG CONVERGENCE RATIO set to .04. We had hoped to get a low WER as a result.

Details:
 * Train configuration
 * Corpus: Switchboard
 * Length: 300hr
 * $CFG_VARNORM = 'no' (variance normalization)
 * $CFG_FINAL_NUM_DENSITIES = 32 (density)
 * $CFG_N_TIED_STATES = 8000 (senones)
 * $CFG_CONVERGENCE_RATIO = 0.04 (convergence ratio)
 * Timeline
 * generateFeats start: 5:33 PM 3/30/16
 * generateFeats end: 6:40 PM 3/30/16
 * Train start: 7:26 PM 3/30/16
 * Train end: 5:11 AM 4/1/16 (34 hours) (91 errors in 001.html)
 * Decode start: 9:58 AM 4/1/16
 * Decode end: 9:17 PM 4/1/16
 * Decode configuration
 * Decoding on: /mnt/main/corpus/switchboard/300hr/test/trans/train.trans (~5 hours)
 * Decoding at: 8000 senones to match the senone count in the train configuration

Results: WER: 35.1%