Speech:Exps 0147

From Openitware
Jump to: navigation, search

Description

Author: Colby Johnson

Date: 04Feb2014

Purpose: The goal of this experiment is to create a Language model and run a Decode using the acoustic model generated in Experiment 0144

Details: This is my first time creating a LM and as it turns out, it is very simple. I will continue to use this experiment to run a decode on the First 5 hour corpus.


Results The LM was created with success.

Decode results on first 5_hr train with default parameters

                    SYSTEM SUMMARY PERCENTAGES by SPEAKER
     ,-----------------------------------------------------------------.
     |                            hyp.trans                            |
     |-----------------------------------------------------------------|
     | SPKR    | # Snt # Wrd | Corr    Sub    Del    Ins    Err  S.Err |
     |=================================================================|
     | Sum/Avg | 2278  35878 | 68.7   23.8    7.6   17.5   48.8   97.7 |
     |=================================================================|
     |  Mean   | 51.8  815.4 | 68.9   23.9    7.2   19.3   50.4   97.9 |
     |  S.D.   | 18.4  297.6 |  7.7    6.2    2.8    9.8   12.8    3.2 |
     | Median  | 46.0  784.0 | 70.3   22.7    6.7   17.3   47.1  100.0 |
     `-----------------------------------------------------------------'