Fact · Speech recognition
In 2022, von Neumann and colleagues proposed a general framework for computing WER when a recognizer processes recordings with several speakers and outputs several word sequences.
Checked against its source on October 17, 2026
In the source's own words
“We propose a general framework to compute the word error rate (WER) of ASR systems that process recordings containing multiple speakers”
On Word Error Rate Definitions and their Efficient Computation for Multi-Speaker Speech Recognition Systems, Thilo von Neumann et al., arXiv (eess.AS), 2022, Abstract.
Open the original- What it measures
- Multi-speaker WER matters for meeting transcription, where the question is which speaker said each word as well as which words were said.
- What it does not show
- The paper is about definitions and how to compute them efficiently. It does not give a single accuracy figure for any system.