Wires live webcast audio into Claude through four stdio tools: point it at an earnings call URL and it spins up ffmpeg plus faster-whisper to stream back rolling transcripts and a heuristic vocal stress score derived from F0 pitch jitter, hesitation ratio, and voiced-frame stats. Built for sell-side analysts who want real-time signals on how the CFO sounds during Q4 guidance, not just what they said. The stress score is literature-inspired but explicitly not validated against market outcomes, so treat it as a coarse prosody signal. Runs entirely on CPU with int8 Whisper, handles concurrent streams with a configurable cap, and includes a disable_vad flag for low-SNR conference call audio where Silero gets too aggressive.
claude mcp add --transport stdio ykshah1309-live-audio-intelligence-mcp uvx live-audio-intelligence-mcp