sphinx_lm_eval1

Usage

Overview

This program evaluates the perplexity of a text file according to a given language model. The text file is assumed to be in transcript format, i.e. one utterance per line, delimited by <s> and </s>.

-help

Shows the usage of the tool

-lm

model file

-lmctlfn

file listing a set of language models

-lmname

of language model in -lmctlfn to use for all utterances

-logbase

Base in which all log-likelihoods calculated

-lsn

file to evaluate

-lw

Language model weight

-mmap

Use memory-mapped I/O for reading binary LM files

-probdef

definition file for classes in LM

-text

string to evaluate

-uw

Unigram probability weight (interpolated with uniform distribution)

-verbose

Print details of perplexity calculation

-wip

Word insertion probability