Voice Tryouts

Working demos of live voice and speech APIs. Each one runs right here in the browser: allow the microphone, start talking, and the transcript arrives while you speak. No sign-up, no API key of your own.

Audio goes to the OpenAI Realtime API (gpt-live-transcribe) through @xoxo-labs/realtime-transcribe; this site's server mints a short-lived client secret, so no API key is ever exposed to the browser. Sessions on the hosted demo stop themselves after three minutes — start another whenever you like. Each experiment is self-contained and instrumented, so the numbers can be compared.

Live transcription over WebRTC

Press Start and talk: microphone audio streams to OpenAI's Realtime API over a peer connection with gpt-live-transcribe, and the words land as you speak. Every stage of the handshake is timed, down to time-to-first-word.

OpenAI Realtimegpt-live-transcribeWebRTCbenchmark

Voice input for AI Elements

Press the mic in an AI Elements PromptInput and dictate into it. Real API transcription replaces the built-in Web Speech mic, which buys cross-browser behaviour, model choice, a delay dial and pre-roll. There is no chat backend — submitting echoes the text back at you.

AI Elementsgpt-live-transcribews-prerolluseVoiceInput