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

Ever wondered how machines understand speech as we do?

Imagine you're asking your smart assistant to play your favorite song, but it keeps asking for the song's lyrics instead.

The assistant struggles because it can't match spoken words to its stored lyrics perfectly. The WER metric helps us understand how well it gets this right or wrong.

Example

If you asked for "Bohemian Rhapsody" and the assistant said "Bohemian Rhapsody lyrics," there's a mismatch.

Remember this

WER measures the assistant's accuracy in recognizing your spoken words.

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?

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

Ever wondered how computers understand words?

weight tying does in language models: shares embedding and output projection matrices

Ever wonder how machines understand the sequence of words in a sentence?

ring attention does: distributes long sequences across multiple devices

How can a machine understand and generate human language?

QLoRA adds

Ever wondered how computers can understand and mimic human speech?

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

Swipe through 100 ML concepts daily

Open Pocket Polymath