========= pysrilm

Python Interface to SRILM

Copyright (c) Will Roberts 9 June, 2015

License: The source code included in this package is licensed under the MIT License (see LICENSE.txt). To install, you will also need a copy of the SRILM toolkit_, for which you will need a license from SRI.

.. _SRILM toolkit: http://www.speech.sri.com/projects/srilm/

Getting Started

You can load a n-gram model like this (note that because SRILM understands zlib, you can pass gzipped files to the read method)::

import srilm
ngrams = srilm.Ngrams(3)
ngrams.read('ngram-counts-eu.tsv.gz')

Look up the frequency of some n-gram in the model::

ngram = ['red', 'flower']
freq = ngrams.find_count(ngram)

Iterate over all the n-grams of a certain order in a particular model::

for (bigram, count) in ngrams.ngrams(2):
    print ' '.join(bigram) + ':', count

Install

Put the SRILM source in the srilm directory in this directory and do python setup.py install or python setup.py build_ext.

Development

For faster turnarounds when hacking on the source code, you can do CC="ccache gcc" CXX="ccache g++" python setup.py build_ext, as noted in this StackOverflow answer <http://stackoverflow.com/a/13176803/1062499>_.