Real-Time AI Latency Budget Worksheet

One-page planner: the fixed pipeline stages with their typical millisecond costs (capture, codec, network, relay/SFU, jitter buffer, decode, render), the AI-inference reference numbers (on-device segmentation, streaming ASR, streaming TTS, voice-to-voice), the speed-of-light geography tax (~10 ms RTT per 1,000 km), the ITU-T G.114 perception zones, and the on-device / edge / cloud placement rule.