| Mode | Language Setup | Spoken Output | Typical Credit Pace |
|---|---|---|---|
| Translate Both Ways | Select exact Language A & Language B (58 languages) | Reads each turn aloud in the other speaker's language | ~85 credits / minute (~23 min per 2,000 monthly credits) |
| One-Way Translation Session | Target language set; source fixed or auto-detected | Live translated transcript (optional voice) | Standard live audio/text session metering |
| Karaoke Readback Practice | On-device Apple Speech word highlighting | You read aloud; words highlight in real time | On-device highlighting uses zero cloud audio |
Why Fixing Both Languages Beats Auto-Detect at a Table
Short conversational acknowledgments—names, numbers, 'okay', brief cognates—often sound ambiguous to automatic language detectors. In a two-person dialogue, a single misdetected word can translate a sentence back into the speaker's own language.
GPT Realtime Assistant solves this when you toggle on Translate Both Ways: you select the exact pair from 58 supported languages before tapping Start. Because two translation pipelines stand ready simultaneously, each person speaks naturally in turn and hears their sentence rendered into the partner's language.
Setting Up the Table for Clean Turn-Taking
Good acoustic placement matters more than speaking loudly:
- Open GPT Realtime Assistant on iPhone, iPad, or Mac, start a Translation session, enable Translate Both Ways, and pick both languages.
- Turn on Speak Translations so neither person has to lean over to squint at every line.
- Lay the iPhone or iPad flat on the table within 40–60 cm of both speakers, away from espresso machines, street traffic, or restaurant speakers.
- Speak in complete, natural sentences of one to three clauses, then pause briefly so semantic voice detection finishes the turn and speaks the translation before the other person replies.
Keep the App Open on iOS and Obtain Consent
On iPhone and iPad, keep GPT Realtime Assistant open on screen throughout the live session so microphone streaming stays uninterrupted. Always ask the other person's permission before starting a live session, and verify important dates, dosages, prices, or legal numbers visually on screen.
Pre-Session Checklist
Saved locally in your browser so you can check your setup before starting.
Frequently Asked Questions
Why does two-way translation use around 85 credits per minute?
Bidirectional mode maintains two simultaneous live translation streams over OpenAI's Realtime API and synthesizes spoken audio back out. The 2,000 AI credits included with the $9.99/month subscription cover roughly 23 minutes of continuous two-way voice interpretation, with 1,000 and 5,000 credit top-ups available anytime.
Are our translated audio recordings stored on Arte Soft's servers?
No. Audio streams directly from your device to OpenAI's Realtime API using a short-lived session token. Arte Soft's backend records credit usage and session titles for billing, never your audio or transcripts—transcripts remain locally on your device.
Continue Exploring
Talking to a Live Voice AI Assistant Hands-Free—and Interrupting Naturally
Brainstorm, rehearse, and work through complex problems out loud without tapping push-to-talk. How semantic turn detection and barge-in interruption work.
Read session guide →
Asking a Voice AI About Screenshots, Photos, and Shared Mac Screens Mid-Call
Attach screenshots, camera photos, or captured Mac windows to a live voice session so GPT Realtime Assistant answers with your exact diagram, error log, or slide in view.
Read session guide →