ssss1029

10/2/2018

Goal:

Use Kaldi on WSJ data to train and decode with traditional HMM-GMM monophone model, then a triphone HMM-GMM model, and then a simple HMM-DNN model.

Background

• GMM & E-M Algorithm Gaussian mixture models and the EM algorithm: https://people.csail.mit.edu/rameshvs/content/gmm-em.pdf
• Jurafsky & Martin (Chapter 6, 7, and 9): http://stp.lingfil.uu.se/~santinim/ml/2014/JurafskyMartinSpeechAndLanguageProcessing2ed_draft%202007.pdf
• HMM-GMM Aoustic Models for Speech Reognition: http://www1.icsi.berkeley.edu/~arlo/publications/faria_cs281a_proj.pdf

ssss1029

\documentclass{article}

% if you need to pass options to natbib, use, e.g.:
%     \PassOptionsToPackage{numbers, compress}{natbib}
% before loading neurips_2019

% ready for submission
% \usepackage{neurips_2019}

% to compile a preprint version, e.g., for submission to arXiv, add add the

ssss1029

Epoch: [0][ 7630/21352] Time  1.768 ( 2.713)    Data  0.000 ( 0.067)    Loss 6.2701e-01 (7.4326e-01)    Acc@1  60.00 ( 61.05)   Acc@5  86.67 ( 80.43)                         [1868/1868]
Epoch: [0][ 7640/21352] Time  2.724 ( 2.713)    Data  0.001 ( 0.075)    Loss 6.2641e-01 (7.4508e-01)    Acc@1  61.67 ( 61.01)   Acc@5  88.33 ( 80.35)
Epoch: [0][ 7640/21352] Time  2.691 ( 2.713)    Data  0.001 ( 0.067)    Loss 6.2034e-01 (7.4316e-01)    Acc@1  65.00 ( 61.05)   Acc@5  86.67 ( 80.44)
Epoch: [0][ 7650/21352] Time  3.767 ( 2.713)    Data  0.001 ( 0.075)    Loss 7.1818e-01 (7.4511e-01)    Acc@1  61.67 ( 61.01)   Acc@5  76.67 ( 80.35)
Epoch: [0][ 7650/21352] Time  3.647 ( 2.713)    Data  0.000 ( 0.067)    Loss 6.9878e-01 (7.4314e-01)    Acc@1  60.00 ( 61.05)   Acc@5  80.00 ( 80.44)
Epoch: [0][ 7660/21352] Time  1.645 ( 2.713)    Data  0.000 ( 0.075)    Loss 9.9575e-01 (7.4507e-01)    Acc@1  51.67 ( 61.01)   Acc@5  71.67 ( 80.35)
Epoch: [0][ 7660/21352] Time  1.768 ( 2.713)    Data  0.000 ( 0.067)    Loss 5.8892e-01