Word error rate

Encyclopedia : W : WO : WOR : Word error rate

Word error rate (WER) is a common metric of measuring the performance of a speech recognition system.

The general difficulty of measuring the performance lies on the fact that the recognized word sequence could have different length from the reference word sequence (supposingly the correct one). The WER is derived from the Levenshtein distance, working at word level instead of character.

This problem is solved by first aligning the recognized word sequence with the reference sequence using dynamic string alignment.

Word error rate could then be computed as:

[ WER = \frac ]

However, when reporting the performance of a speech recognition system, the word recognition rate (WRR) is used instead:

[ WRR = 1 - WER = \frac = \frac ]

where

N is the number of words in the reference,
S is the number of substitutions,
D is the number of the deletions,
I is the number of the insertions,
H is the number of correctly recognised words.

References

[McCowan et al. 2005: On the Use of Information Retrieval Measures for Speech Recognition Evaluation]

From Wikipedia, the Free Encyclopedia. Original article here. Support Wikipedia by contributing or donating.
All text is available under the terms of the GNU Free Documentation License See Wikipedia Copyrights for details.