Live Voice Mode on VEGA: Real-Time Conversations with GPT-Live 1
We launched Live, a new mode on VEGA for real-time voice conversations with AI. Powered by OpenAI's GPT-Live 1 (openai/gpt-live-1), Live mode lets you speak naturally in your native language and hear spoken replies immediately, without typing prompts or waiting for a separate speech-to-text pipeline.
What makes Live different from voice dictation
Voice dictation turns your speech into text inside the input box so you can send a regular written message. Live is a full-duplex voice session: the model listens and speaks over a continuous 24 kHz PCM audio stream.
- Speak in your native language: GPT-Live 1 detects the language you speak and answers in the same language with natural pacing and intonation.
- Smooth interruption (barge-in): If you start speaking while VEGA is still talking, playback stops right away and the model shifts back to listening—just like a real phone call.
- Live equalizer in the chat: A 20-bar audio visualizer reacts in real time both to your microphone level and to VEGA's voice output, so you always see whether the session is listening, thinking, or speaking.
- Live subtitles and chat history: Spoken turns are transcribed as the conversation happens and saved directly into the chat thread for easy reading or export after the call.
How it works under the hood
When you press Start conversation in the Live tool group:
- VEGA mints a single-use client secret on the server and opens a direct WebSocket session to the AI Gateway Live endpoint.
- Your browser captures microphone audio through an
AudioWorklet, resamples it to 24 kHz mono PCM16, and streams it with minimal latency. - Incoming audio deltas from
openai/gpt-live-1are scheduled frame-by-frame in the Web Audio API, while live transcripts stream into the chat interface. - You can mute or unmute your microphone at any moment, or end the call to finalize the session cleanly.
Per-second billing
Live sessions are billed by actual duration in seconds from your VEGA deposit, with no rounding up to whole minutes. When you end the call, the hold is settled against the exact session time reported by the provider.
Try Live on VEGA
Open vega.chat, select Live in the tool selector bar, and click Start conversation. Put on your headphones or use your device microphone and talk to VEGA in any language.