Before the event
- Event glossary - Up to 200 term pairs per language pair (names, product terms, agreed renderings), enforced in translation with case-insensitive matching, per direction.
- Event terminology - Key phrases per language prime both speech recognition (so names and jargon are heard correctly) and translation (so they are rendered consistently).
- Register and audience - Formal, informal or neutral register, a description of the audience, and free-form instructions to the translation engine, set per event.
- Rehearsal - A full technical rehearsal with your team on the real setup, producing a real latency and quality report before the doors open.
During the event
- Phrase-aware processing - The engine waits for phrases to stabilise before translating, so listeners hear complete, settled sentences instead of stuttering fragments.
- Context-aware translation - Each language channel keeps a rolling context of the talk and learns the speaker's terminology as it goes, keeping names and terms consistent.
- Automatic failover - If the primary translation engine has an outage mid-session, the channel fails over to a backup engine automatically.
- Live monitoring - The operator console shows live transcripts and pipeline health; our engineer watches the same data during supported events.
- Human in the loop - Any channel can be assigned to a professional human interpreter instead of AI - for the languages or sessions where you want a person in the chain.
After the event
- Per-language latency report - Median and p99, stage by stage, attached to every session.
- Transcripts and translations - For every session, plus per-language MP3 recordings when enabled - your material for review and reuse.