Word/character error rate (speech-BCI metrics)
The standard accuracy metric borrowed from automatic speech recognition: the edit (Levenshtein) distance between decoded and target text, normalized by the target length, counting substitutions, insertions and deletions. It is reported at the word level (WER) or at the character or phoneme level (CER, PER).
Because a language model reshapes and often reduces errors, raw phoneme error rate and post-language-model word error rate for the same system can differ by an order of magnitude.
WER is interpretable only alongside the vocabulary size, whether the corpus is closed or open, whether decoding was real-time or offline, and whether an autocorrect or language model was applied; a headline WER without this context cannot be compared across studies.