← All Articles

OpenClaw Talk Mode Improvements: Configurable Silence Detection

A flurry of commits hardens voice conversation handling across iOS and macOS — configurable silence timeouts, shared config parsing, and better edge case handling.

Why This Matters

Voice interfaces are delicate. The difference between "I'm thinking" and "I'm done talking" is subjective, varies by speaker, and depends on context. When an AI assistant cuts you off mid-thought, it's jarring. When it waits too long after you've finished, it feels unresponsive.

This batch of improvements gives users control over that threshold and makes the behavior consistent across Apple platforms.

What Changed

Configurable Silence Timeout

The core change from contributor @danodoesdesign adds a configurable silence timeout to Talk mode. Previously, the timeout was hardcoded — now users can adjust it in settings:

{
  "talk": {
    "silence_timeout_ms": 1500,
    "sensitivity": "normal"
  }
}

Values range from 500ms (aggressive, for quick-paced conversations) to 3000ms (patient, for thoughtful speakers or those with speech differences). The default remains 1200ms, unchanged from previous behavior.

Hardened Parsing

Peter Steinberger's follow-up commits add defensive parsing for the silence timeout config:

• Invalid values fall back to defaults rather than crashing
• Negative values are rejected
• Extremely large values are capped at 10 seconds
• Type mismatches (string instead of number) are handled gracefully

Shared Apple Talk Config

The refactor commits centralize Talk config parsing into a shared module used by both iOS and macOS. This means:

• Feature parity — both platforms behave identically
• Single source of truth for validation logic
• Easier maintenance for future Talk mode features

Previously, each platform had its own parsing code with subtle differences. Users reported that the same config produced different behavior on their iPhone vs. Mac — that's now fixed.

Canonical Config Payload

A new canonical payload structure ensures that Talk config is serialized consistently when syncing between devices or exporting settings:

{
  "talk": {
    "version": 1,
    "silence_timeout_ms": 1500,
    "sensitivity": "normal",
    "vad_model": "whisper-vad",
    "auto_punctuate": true
  }
}

The version field enables forward compatibility — future Talk config changes can migrate older configs gracefully.

Test Coverage

The PR includes new test coverage for edge cases:

• Empty config objects
• Missing required fields
• Boundary values (0ms, negative, maximum)
• Invalid types (arrays, nested objects where scalars expected)
• Unicode and special characters in string fields

This is part of OpenClaw's broader push toward defensive coding, following the security hardening initiatives from earlier this month.

User-Facing Impact

For most users, nothing changes — the defaults are unchanged. Power users gain:

1. Adjustable pacing — Match the AI's listening behavior to your speaking style
2. Accessibility improvements — Longer timeouts for users who need more time between phrases
3. Cross-device consistency — Your Talk config now works identically on all Apple devices

What's Next

These improvements lay groundwork for more sophisticated Talk mode features:

• Adaptive silence detection — ML-based detection that learns your speaking patterns
• Context-aware timeouts — Longer pauses when reading aloud, shorter for quick commands
• Per-conversation settings — Different profiles for different use cases

The commits are available starting with 6ff7e8f through the day's refactor series.