SYNOPSIS

DESCRIPTION

Samples can be either in plain format (one word per line followed by phonetic transcription) or Bliss XML Lexicon format.

OPTIONS

--version

show program's version number and exit

-h, --help

show this help message and exit

-p FILE, --profile=,FILE/

Profile execution time and store result in FILE

-R, --resource-usage

Report resource usage execution time

-Y, --psyco

Use Psyco to speed up execution

--tempdir=,PATH/

store temporary files in PATH

-t FILE, --train=,FILE/

read training sample from FILE

-d FILE / N%, --devel=,FILE/ / N%

read held-out training sample from FILE or use N% of the training data

-x FILE, --test=,FILE/

read test sample from FILE

--checkpoint

save state of training in regular time intervals. The name of the checkpoint file is derived from --writemodel.

--resume-from-checkpoint=,FILE/

load checkpoint FILE and continue training

-T, --transpose

Transpose model, i.e. do phoneme-to-grapheme conversion

-m FILE, --model=,FILE/

read model from FILE

-n FILE, --write-model=,FILE/

write model to FILE

--continuous-test

report error rates on development and test set in each iteration

-S, --self-test

apply model to development set and report error rates

-s l1,l2,r1,r2, --size-constraints=,l1/,l2,r1,r2

multigrams must have l1 ... l2 left-symbols and r1 ... r2 right-symbols

-E, --no-emergence

do not allow new joint-multigrams to be added to the model

--viterbi

estimate model using maximum approximation rather than true EM

-r, --ramp-up

ramp up the model

-W, --wipe-out

wipe out probabilities, retain only model structure

-C, --initialize-with-counts

initialize probabilities estimation by counting how many times every graphone occurs in the training set, disregarding possible overlaps

-i MINITERATIONS, --min-iterations=,MINITERATIONS/

minimum number of EM iterations during training

-I MAXITERATIONS, --max-iterations=,MAXITERATIONS/

maximum number of EM iterations during training

--eager-discount-adjustment

re-adjust discounts in each iteration

--fixed-discount=,D/

set discount to D and keep it fixed

-e ENC, --encoding=,ENC/

use character set encoding ENC

-P, --phoneme-to-phoneme

train/apply a phoneme-to-phoneme converter

--test-segmental

evaluate only at segmental level, i.e. do not count syllable boundaries and stress marks

-B FILE, --result=,FILE/

store test result in table FILE (for use with bootlog or R)

-a FILE, --apply=,FILE/

apply grapheme-to-phoneme conversion to words read from FILE

-V Q, --variants-mass=,Q/

generate pronunciation variants until \sum_i p(var_i) >= Q (only effective with --apply)

--variants-number=,N/

generate up to N pronunciation variants (only effective with --apply)

-f FILE, --fake=,FILE/

use a translation memory (read from sample FILE) instead of a genuine model (use in combination with -x to evaluate two files against each other)

--stack-limit=,N/

limit size of search stack to N elements