Connector Web

Connector

Email Password

Display name Email Password Confirm password

Logged in as:

User

User

Calling...

Live translation

Establishing connection...

Mic 100%

Incoming 100%

Original 40%

Users

Incoming call

User

Call history

User

Account

Account Settings

Change photo

Email Name Country Language

Permissions

Microphone Unknown

Speaker output Default device

Call ringtone Default Upload

Realtime translation

AI Settings

These settings only affect how your own outgoing speech gets translated for whoever you're talking to. The other person's speech is translated using their own AI Settings, independently - changing yours never affects their calls, and vice versa.

Realtime model

This model currently fails on OpenAI's side (inference_not_found_error) - translated audio will not work.

gpt-realtime-translate is a dedicated mechanical translation endpoint - it has no prompt/instructions and no configurable speech detector at all, so the voice, prompt, VAD sensitivity, cutoff delay, prefix padding, and interrupt/auto-start switches below have no effect and are disabled while it's selected. Only the model choice and each user's own language matter for it.

Translation prompt

Leave blank to use the built-in default prompt (recommended). If you write your own, {source} and {target} are replaced with the actual language names at call time.

Voice VAD sensitivity (0 = permissive, 1 = strict) - 0.5 Audio stream cutoff delay (ms) - 500ms

How long a pause must last before the system decides you've finished talking and commits the response. Too low (e.g. under ~300ms) can cut off natural pauses mid-sentence and cancel the translation before any audio comes back - this was confirmed causing silent translated calls.

Prefix padding (ms) - 300ms Allow new speech to interrupt (barge-in) an in-progress translation Automatically start translating once speech stops (server-driven) Show live subtitles of the translated speech