sphinx-eval

This project is a set of bash scripts that helps to evaluate CMU sphinx ASR system. These scripts can be used after you have the decoder output.

collect_wer.sh

This script extracts WER for the produced transcriptions. WER is extracted from (.align) files using grep command

get_oov.py

This program obtain the list of OOV from ref and hyp files. The benefits is that you can get a list of words that the decoder failed to produce to add it to the system somehow, so it can be recognized in the system.

usage: get_oov.py [-h] --ref REF --hyp HYP --oov OOV

get_oov.sh

This script get OOV words for a transciption with respect the hyp. This script uses get_oov.py to get OOV words for a bunch of files.

get_diacritcs_from_mishkal

The script use miskal utility to add diacritics to Arabic texts.

For more information about mishkal, please refer to https://tahadz.com/mishkal and https://github.com/linuxscout/mishkal

ppl.sh

This script gets perplexity (PPL) of a text given a languge model (LM). The script computes PPL for bunch of texts given bunch of LMs.

word_align.sh

This script align words in (hyp and trs) texts using word_align.sh script