word error rate (WER) measures: edit distance between predicted and reference transcriptions

WER measures the percentage of errors in transcription

Word error rate (WER) quantifies the accuracy of speech recognition systems by comparing the predicted transcription to a reference transcription. It is expressed as a percentage, indicating the proportion of words that were incorrectly transcribed.

Example

If a system transcribes "The quick brown fox jumps over the lazy dog" as "The quick brown fox jumps over the lazy cat," the WER would be 20% because 1 out of 5 words is incorrect.

Remember this

Understanding WER helps developers improve speech recognition systems by identifying and reducing transcription errors.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

Evaluation of machine translation

Can we truly measure how good a machine translation is?

BLEU vs ROUGE: BLEU measures precision of n-grams, ROUGE measures recall

BLEU measures precision of n-grams, ROUGE measures recall

WordPiece tokenization does: similar to BPE but uses likelihood instead of frequency

WordPiece tokenization splits words into subwords based on token likelihood rather than frequency

the vocabulary size matters: larger vocab = shorter sequences but more parameters

Larger vocab leads to shorter sequences but more parameters

Brier score

Brier score measures mean squared error of probability predictions

300-dim word2vec encodes: trained on word co-occurrence with skip-gram window

Ever wondered how computers understand words?

Swipe through 100 ML concepts daily

Open Pocket Polymath