Word Error Rate (WER)

Word error rate is the standard measure of how accurate a speech-recognition system is — the percentage of words it gets wrong against a correct transcript — and it’s one of the few numbers worth asking a voice AI vendor for directly.

A demo call in a quiet room with a clear speaker rarely reflects the real conditions a system will face — background noise, regional accents, a code- switched sentence that moves between two languages mid-thought, a bad mobile signal. Word error rate on those harder, realistic conditions is a far more honest signal of how a system will actually behave in production than a polished demo.

There’s no single number that applies everywhere — accuracy varies by language, accent and audio quality — which is exactly why it’s worth asking a vendor to demonstrate it on a call that sounds like a real customer, in the language mix a business actually gets, rather than accepting a headline accuracy figure at face value.

More on evaluating a vendor, on how to choose between vendors.