[szhaomsft]

Summary

• Add an editable 0–100% both-parties-silent ratio, defaulting to 20%.
• Reduce speech/audio usage only: the default 45-second allocation per side becomes 36 seconds, with 360 input and 720 output audio tokens.
• Apply the reduction to fresh and replayed audio, STT, and TTS, while preserving text, prompts, reasoning, turn counts, GPT Live session-minute charges, and full-call per-minute normalization.
• Show the adjustment in the summary and CSV, document the illustrative 10–30% planning range, and explain how to avoid double-counting measured silence.

Validation

• All 23 calculator tests pass, including zero/full silence, intermediate ratios across pipeline/hosting modes, invalid ratios, unchanged unit prices, and unaffected non-audio costs.
• Browser verification covers initial/default usage, zero-silence baseline, unchanged GPT Live costs, validation, reset, and CSV export.
• JavaScript syntax and whitespace checks pass.

Deployment

The user authorized merging after checks pass to publish through the existing GitHub Pages workflow.