Custom vocabulary
A per-event list of domain-specific terms (product names, technical jargon, speaker names) supplied to an ASR system to improve transcription accuracy on those terms.
ASR models are trained on general speech and don't reliably recognize words that aren't in their training distribution: brand names ('Kubernetes'), internal acronyms ('OKR', 'NPS'), uncommon proper nouns ('Eze Ezzejjari'), or technical terms specific to a niche.
Custom vocabulary lets an event configure these terms ahead of the broadcast, dramatically improving accuracy on the first transcript pass. A 30-term custom vocabulary typically takes 2-3 minutes to assemble and lifts technical-content accuracy noticeably.
Most modern captioning platforms expose custom vocabulary as a per-event setting. The terms supplied are typically biased into the ASR model rather than treated as a hard substitution rule, so the model still produces plausible alternatives if it mishears.