Costa

Costa provides statistics about corpora, lexicons, and language models.

Parameters:

evaluate-recordings : feed the audio files specified in the corpus to the flow network specified by feature-extraction. This is useful if you want to check that all audio files are present and readable.

lexical-statistics : print statistics about the lexicon

lm-statistics : print statistics about the language model

feature-extraction : see feature extraction

corpus : see Common component configuration

lexicon : see Common component configuration

lm : see SpeechRecognizer

Channels:

recordings : recording names

lexicon.dump : dump the lexicon

Example Configuration

[*]
evaluate-recordings             = yes
lexical-statistics              = yes
lm-statistics                   = no

# define the lexicon used
[*.lexicon]
file                            = ../lexicon/training.lexicon.gz

# do not load the language model, just simple zerogram statistics
[*.lm]
type                            = zerogram

# define the corpus
[*.corpus]
file                            = corpus.gz
audio-dir                       = /path/to/audio/files

# Flow file with a single source node to open and scan the audio file
[*.feature-extraction]
file                            = config/test-audio.flow

# Channels
[*]
log.channel                     = log-channel
warning.channel                 = log-channel, stderr
error.channel                   = log-channel, stderr
statistics.channel              = log-channel
configuration.channel           = log-channel
unbuffered                      = true
log-channel.file                = $(log-file)