Preprint
Retrieval Augmented Correction of Named Entity Speech Recognition Errors
Ernest Pusateri and et al. (nine authors in total), arXiv, 2024
Read the originalRead on October 16, 2026
- What it is
- A preprint from Apple researchers on the errors that speech recognition makes with names of people, places and things (named entities). It proposes a correction step that uses retrieved lists of entity names.
- How much weight it carries
- A preprint, not peer-reviewed in the version we read. Its test sets are synthetic and built around voice assistant requests, so the error rates describe those tests, not long-form dictation.
Facts taken from this source
- The 2024 paper says end-to-end speech recognition systems still have a significant error rate for entity names that appear rarely in their training data.
“these systems still have a significant error rate for entity names which appear infrequently in their training data” (Abstract)
- Apple researchers report that end-to-end speech recognition still has a significant error rate for entity names that appear infrequently in its training data.
“these systems still have a significant error rate for entity names which appear infrequently in their training data” (Abstract)
Pages that rest on this source
- AI Voice Agents: How They Work and Where They FailGuide
- Writing a Book by Dictation: Software and WorkflowGuide
- How Accurate Is Speech Recognition? The Real NumbersGuide
- Dictation for Academics, Researchers and ThesesGuide
- Can you dictate a novel?Question
- How do you dictate citations and references?Question
- Why does speech recognition get names and jargon wrong?Question