foundry-voice-live

What problem does it solve?

Builds real-time, low-latency voice agents on Azure AI Foundry by guiding developers to migrate from Azure OpenAI Realtime to Azure Voice Live (GA) with the correct session configuration, routing, and UI plumbing.

Core Features & Use Cases

• Three-rung migration ladder: Provides copy-paste-ready connection differences for Realtime, Voice Live, and Voice Live routed through a Foundry Agent.
• Voice Live session tuning: Sets azure_semantic_vad, server echo cancellation, azure deep noise suppression, and low-latency transcription with Azure Fast Transcription.
• Agent routing via extra_query: Enables Foundry Agent activation by passing agent-id, agent-project-name, and agent-access-token on the Voice Live WebSocket.
• Benchmarking latency with TTFA/TTFT: Defines a scenario matrix and measures request-to-audio and request-to-first-token latency patterns.
• Gradio + FastRTC WebRTC UI integration: Shows how to wire a real-time browser microphone/speaker pipeline to the same Realtime event schema.

Quick Start

Use the foundry-voice-live skill to wire your WebRTC/Gradio demo to Azure Voice Live by switching MODE to demo and supplying AZURE_OPENAI_ENDPOINT and AZURE_VOICELIVE_ENDPOINT in your .env.