OpenAI's low-latency voice AI ensures conversational interactions occur at the speed of speech, minimizing awkward pauses and interruptions. This capability benefits ChatGPT voice, developers using the Realtime API, and agents engaged in interactive workflows.

GitHub repository mahimairaja/voiceai provides a curated learning path for developers to build real-time voice AI agents, covering the process from speech-to-text (STT) to production telephony. The modern voice AI stack includes a real-time transport layer, a streaming pipeline of speech-to-text, large language models (LLM), and text-to-speech technologies, along with a turn-taking model for managing agent interactions.

RCLI is an on-device voice AI for macOS that allows users to interact with their Mac and query documents without requiring cloud services. It features a complete STT, LLM, and TTS pipeline running natively on Apple Silicon, with 38 macOS actions available via voice and sub-200ms end-to-end latency.