Streaming speech recognition and response generation during live calls - > Real-time Speech agent

Intent-based journey orchestration - > NLU in Copilot Studio

A real-time voice agent is designed for low-latency conversational processing in which spoken input and generated responses are handled continuously during the live interaction. Microsoft distinguishes real-time voice agents from basic voice agents: real-time agents use a speech-to-speech real-time model for natural, low-latency conversations, while basic voice agents typically convert speech to text, process the resulting text, and then synthesize speech.

Consequently, Real-time Speech agent is the appropriate component for streaming speech recognition and response generation during a live call. It addresses the requirement for fluid, interactive speech rather than merely supplying a standalone transcription service.

For intent-driven orchestration, Copilot Studio's Natural Language Understanding capability interprets customer language and determines which predefined topic or conversational path should execute. Microsoft documents NLU, NLU+, and classic orchestration as mechanisms for deterministic intent-driven dialog behavior, while generative orchestration dynamically chooses topics, actions, and knowledge.

Voice routing configuration operates at the Contact Center work-distribution layer rather than performing conversational intent understanding. Azure AI Speech provides speech services but does not itself represent the Copilot Studio orchestration component required here.

Study Guide reference: Configure agents and AI capabilities - > Configure voice agents, real-time voice, NLU, and IVR orchestration.