Speech:Exps 0272 001

Description
Author: Michael Jonas

Date: 4-2-2015

Purpose: Copied 0276/001 to look at results and do some scoring

Details:

All of 256 hours of Switchboard corpus was used to train acoustic models.

Of the 256 hour training corpus this decode was run on approximately the first 19 hours.

Results:

,-.     |                            hyp.trans                            | |-|     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err | ...     |-+-+-|      | Sum/Avg |18806  268452| 53.0   37.1    9.9   14.5   61.5   98.4 | |=================================================================|     |  Mean   | 64.0  913.1 | 53.5   37.1    9.4   15.7   62.2   98.5 | | S.D.   | 25.5  361.4 |  8.7    7.2    3.3    7.1   10.9    2.4 | | Median | 63.0  915.5 | 53.7   37.2    8.8   14.4   61.7  100.0 | `-'

So a WER of 61.5%.