The pipeline
Everything starts with a single audio feed and ends on every listener's device, in parallel for every language:
- 1
Stage audio
One feed from your sound console - any number of microphones, panels and playback sources, mixed by your engineer exactly as today.
- 2
Speech recognition
Streaming, phrase-aware recognition waits for phrases to stabilise, so the translator sees complete, settled sentences instead of fragments.
- 3
AI translation
Context-aware translation keeps a rolling context of the talk and honours your event glossary and terminology.
- 4
Neural voice
A natural neural voice speaks each target language in parallel, one channel per language.
- 5
Delivery
Audio and live captions reach every listener over WebRTC - phones in the hall, remote browsers, venue screens and production tools.
End-to-end delay is typically about 5 seconds from the spoken phrase to the interpreted voice. The platform measures this continuously: after every session you receive a per-language latency report (median and p99), so quality is verified with data, not impressions.
Engines and voices
Interpretwise runs several speech-recognition, translation and voice engines behind the scenes and picks the best combination for each language pair, with automatic failover between engines mid-session. Voices come in two tiers:
- Premium voices - The most natural option, recommended for stage content.
- Standard voices - A large catalogue covering the full language range at a lower rate.
Humans in the loop
The same event can mix AI interpretation and professional human interpreters channel by channel: for example AI for six languages and a human interpreter for the seventh. Interpreters work from anywhere through a browser-based console - no installs, protected by a per-event access code.
What attendees experience
Attendees scan a QR code, pick a language and listen on their own phone - with live captions in the same language. No app, no account, no headsets to distribute. See Audience without headsets for the full attendee flow.
