Build an Agora Transcription Bot with AssemblyAI Universal-3 Pro

Build an Agora Transcription Bot with AssemblyAI Universal-3 Pro This tutorial walks through building a real-time transcription bot in Python that joins an Agora channel as a silent observer, captures each participant's audio as raw PCM frames, and streams it to AssemblyAI Universal-3 Pro Streaming for speaker-aware transcripts. The full source is available at github.com/kelseyefoster/voice-agent-agora-universal-3-pro. Why This Stack? Agora's Python Server SDK lets a server-side bot join channels, subscribe to participant audio as raw PCM frames, and optionally publish audio back — without any browser or mobile client. This PCM stream format aligns directly with what AssemblyAI Universal-3 Pro Streaming expects, making the integration unusually clean. Metric AssemblyAI Universal-3 Pro Agora Built-in STT P50 latency 307ms ~600–900ms Word Error Rate 8.9% ~14–18% Speaker diarization ✅ Real-time ❌ Languages 99+ Limited Prerequisites Python 3.9+ Agora Console account — App ID and App Certif