3

I am looking for a library, script or program that can evaluate the output of a speech recognition system. The output of the speech recognition system is a simple text file, and I have the gold output in the same format.

I have crossposted the question at:

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Franck Dernoncourt
  • 42,093
  • 30
  • 155
  • 271
  • What do you mean by "evaluate"? You just want a measure of how accurate the output is relative to the gold, like an F-score, perhaps [https://en.wikipedia.org/wiki/F1_score](https://en.wikipedia.org/wiki/F1_score)? This is relatively easy to write, e.g. in R. Some examples as part of other software: [http://homepages.inf.ed.ac.uk/sgwater/resources.html](http://homepages.inf.ed.ac.uk/sgwater/resources.html) or [http://web.science.mq.edu.au/~mjohnson/Software.htm](http://web.science.mq.edu.au/~mjohnson/Software.htm) – Rose Hartman May 08 '17 at 02:48
  • @RoseHartman thanks, I mostly had in mind the [word error rate](https://en.wikipedia.org/wiki/Word_error_rate). It's not that straightforward as one needs to align the predicted text with the gold text first. I see several scripts computing the word error rate, such as https://github.com/zszyellow/WER-in-python or https://martin-thoma.com/word-error-rate-calculation/, but I'm unsure which ones are reliable. – Franck Dernoncourt May 08 '17 at 04:24

0 Answers0