Realtime Voice & Live Translation

Overview

ToRun supports realtime, bidirectional voice — you speak, the assistant listens and responds in audio and text over a live session — and live translation, where speech is translated as it happens. These run on the realtime capability (bidirectional audio + text) and are billed per second of session time.

Capabilities involved

• Realtime (audio + text) — a live WebSocket-style session with a capable provider model for natural back-and-forth voice conversation.
• Text-to-speech (TTS) — synthesized speech output across 29 languages.
• Speech-to-text (STT) — transcription, including speaker diarization (who said what).
• Live translation — speech translated in real time between languages.

Starting a realtime session

1. Open the voice/realtime surface.
2. Grant microphone access.
3. Speak naturally — the assistant transcribes, responds, and (in voice mode) speaks back.
4. End the session when done. The session is tracked as a voice session entity.

For live translation, choose the source and target languages; each party hears the other's speech translated as it streams.

Billing

• Realtime and live-translation usage is metered per second of audio.
• Each metered unit of work produces a BillingRecord with a pricing snapshot, so the cost of a voice session is transparent and auditable.
• Voice/realtime sits behind the Voice/Audio entitlement, which is included from the Pro tier upward.

STT and diarization

Speech-to-text supports speaker diarization, labeling segments by speaker. This is useful for transcribing meetings, interviews, or multi-party calls where you need to attribute lines to specific speakers.

Tips

• Use a headset or a quiet room — clean input audio improves both transcription and translation accuracy.
• Because realtime is metered per second, end sessions you're not actively using.
• For one-off transcription of an existing recording, STT (not a live realtime session) is the cheaper path.