Voice input: ~4 second delay before recording starts (beginning of speech cut off)

Environment

• Claude Code version: v2.1.86 (CLI / Windows Terminal)
• OS: Windows 11 Pro (10.0.26100)
• Microphone: AnkerWork S600 (USB, 1ch/16bit/48kHz, speakerphone with echo cancellation)
• Setting: voiceEnabled: true in settings.json

Steps to reproduce

1. Press and hold spacebar to start voice input
2. Immediately start speaking, e.g.: "Hey, ich habe jetzt neu gestartet. Kannst Du mich hören und mit Katja antworten?"
3. Release spacebar after finishing

Expected behavior

Full sentence is transcribed.

Actual behavior

Only the last few words appear (e.g. "...mit Katja antworten?"). First ~4 seconds of speech are lost.

Root cause hypothesis

The audio capture stream is opened AFTER spacebar press but takes ~4 seconds to initialize. Audio during this init phase is silently dropped, not buffered. A pre-roll buffer or stream pre-init on app start would fix this.

Workaround

Wait ~4 seconds after pressing spacebar before speaking — not practical for daily use.

Additional notes

• Disabling Windows audio enhancements (echo cancellation) did not help
• Issue is reproducible 100% of the time
• USB microphone, not Bluetooth (no additional latency expected)
• Tested with multiple sentence lengths, always first ~4 seconds lost

Suggested fix

Pre-initialize audio stream when Claude Code starts (keep stream open/warm) rather than opening it on spacebar press.