You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The parameter "hypothesis" is understood by me as the result of speech recognition, and "reference" as the actual sentence.
Therefore, the minimum edit distance calculated should be the distance from "hypothesis" to "reference". However, upon testing the existing code, the edit distance is from "reference" to "hypothesis". Could this be a bug?
The text was updated successfully, but these errors were encountered:
You are right, a reference is the gold standard, while the hypothesis is a prediction from the system. You want to measure how many deletions, insertions, or substitutions need to be made FROM the prediction TO the gold standard, in order to know how many mistakes the systems made. The other way around, if you think about it, does not make sense, why should you want to modify the gold standard?
The most important aspect in the calculation is that N in the WER formula S+D+I/N is the length of the reference, and not the hypothesis, which is accomplished by calculating the distance from the hypothesis to the reference.
The parameter "hypothesis" is understood by me as the result of speech recognition, and "reference" as the actual sentence.
Therefore, the minimum edit distance calculated should be the distance from "hypothesis" to "reference". However, upon testing the existing code, the edit distance is from "reference" to "hypothesis". Could this be a bug?
The text was updated successfully, but these errors were encountered: