This directory includes information about the following paper
Yanmin Qian, Maofan Yin, Yongbin You, Kai Yu,
Multi-Task Joint-Learning of Deep Neural Network for Robust Speech Recognition,
ASRU, 2015.

The training and eval data are the standard AURORA4 corpus.
About aurora4
    The aurora4 database contains a) clean wsj0 data (Wall Street Journal)
                                  b) artificially added noise with clean wsj0 data
    for detailed information, please refer to: http://aurora.hsnr.de/aurora-4.html.
    To obtain the data you should contact ELRA, see
    http://catalog.elra.info/index.php?cPath=37_40, and look for aurora4a and aurora4b.
    If you already have the WSJ license from LDC, you should not need any
    additional licenses (but they may want to check that you have a license for WSJ).

About the Wall Street Journal corpus:
    This is a corpus of read sentences from the Wall Street Journal, recorded under clean conditions.
    The vocabulary is quite large. About 80 hours of training data.
    Available from the LDC as either: [ catalog numbers LDC93S6A (WSJ0) and LDC94S13A (WSJ1) ]
    or: [ catalog numbers LDC93S6B (WSJ0) and LDC94S13B (WSJ1) ]
    The latter option is cheaper and includes only the Sennheiser microphone data

About the files in this directory:

We give several systems' .mlf (the ASR output) and .cer (scoring results) files, whose results are the most important in the paper.

File aurora4.test.all.mlf is the reference mlf for the test data, and we could use HResults to do the scoring.

(0) Table1-Baseline: We give the baseline system results shown as the first line of Table1, avg WER 12.4%
the results file: Table1-Baseline/aurora_eval/result.cer
the mlf file: Table1-Baseline/aurora_eval/result.mlf

(1) Table1: We give the most important results shown as the last line of Table1, avg WER 11.2%
the results files: Table1/aurora_eval/results
the mlf files: Table1/aurora_eval/mlfs

(2) Table2: We give the most important results shown as the third line of Table2, avg WER 10.8%
the results files: Table2/aurora_eval/results
the mlf files: Table2/aurora_eval/mlfs

(3) Table3: We give the most important results shown as the last line of Table3, avg WER 10.2%
the results files: Table3/aurora_eval/results
the mlf files: Table3/aurora_eval/mlfs

(4) Table5: We give the most important results shown as the last line of Table5, avg WER 9.7%
the results files: Table5/aurora_eval/results
the mlf files: Table5/aurora_eval/mlfs

P.S. You could reference the HTK tookit (http://htk.eng.cam.ac.uk/) about the "mlf" file format and "HResults" usage.