wavecap-hallucination

Configure WaveCap's Whisper integration to detect and prevent AI hallucinations.

Updated Aug 23, 2026
One-click install
npx skills add https://github.com/TobiasWooldridge/WaveCap --skill wavecap-hallucination
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: wavecap-hallucination
Source: https://github.com/TobiasWooldridge/WaveCap/tree/main/.claude/skills/wavecap-hallucination
Command: npx skills add https://github.com/TobiasWooldridge/WaveCap --skill wavecap-hallucination

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill addresses the issue of Whisper AI generating inaccurate transcriptions, such as gibberish, repeated phrases, or phantom text, especially on silent or low-quality audio.

Core Features & Use Cases

  • Hallucination Detection: Identifies and filters out common hallucinated phrases on low-energy audio.
  • Segment Repetition: Detects and discards excessively repeated phrases.
  • Blank Audio Handling: Emits [BLANK_AUDIO] tokens for segments with minimal speech energy.
  • Use Case: When transcribing a meeting with long silences, this skill prevents the AI from outputting nonsensical phrases like "thank you for watching" during those quiet moments.

Quick Start

Use the wavecap-hallucination skill to configure Whisper to ignore the phrase "thank you for watching" when it appears on silent audio.

Frequently Asked Questions about wavecap-hallucination

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
Why does Whisper AI generate phantom text like "thank you for watching" on silent audio?

Whisper AI generates phantom text on silent audio because it attempts to predict language patterns even without speech. This hallucination occurs on low-energy audio, but filters can detect and discard these common nonsensical phrases.

How do I configure Whisper transcription to stop repeating phrases on low-quality audio?

Configure Whisper transcription to stop repeating phrases by enabling segment repetition detection. This identifies and discards excessively repeated phrases, filtering out non-speech artifacts to optimize transcription accuracy on low-quality audio.

What is blank audio detection in speech recognition transcription?

Blank audio detection in speech recognition identifies segments with minimal speech energy. Instead of generating phantom text, it emits a designated token to indicate silence, preventing AI hallucinations during long quiet moments in audio processing.

Can I filter specific hallucinated phrases from Whisper AI speech recognition output?

Yes, you can filter specific hallucinated phrases from Whisper AI speech recognition output. Silence hallucination phrase management allows you to configure Whisper to ignore specific nonsensical phrases when they appear on low-energy or silent audio.

Does Whisper AI hallucination filtering work for meeting transcriptions with long silences?

Whisper AI hallucination filtering works effectively for meeting transcriptions with long silences. It prevents the AI from outputting gibberish or repeated phrases during quiet moments, ensuring accurate speech recognition for prolonged silent audio segments.

What are the limitations of using hallucination detection for audio processing?

Hallucination detection for audio processing relies on identifying common phrases and minimal speech energy, meaning it targets known non-speech artifacts. It optimizes transcription accuracy by filtering gibberish but is limited to configured hallucination parameters and silence thresholds.