Skip to content

Conversation

@safayavatsal
Copy link

  • Created whisper/enhancements module for enhanced functionality
  • Implemented HallucinationDetector with multi-method detection:
    • Pattern-based detection (YouTube artifacts, repetitive phrases)
    • Statistical analysis (compression ratios, log probabilities)
    • Repetition analysis (looping behavior detection)
    • Temporal analysis (silence-based detection)
  • Added ConfidenceScorer for comprehensive transcription quality assessment
  • Enhanced transcribe() function with new parameters:
    • enhanced_hallucination_detection: Enable advanced detection
    • hallucination_detection_language: Language-specific patterns
    • strict_hallucination_filtering: Strict vs permissive filtering
    • confidence_threshold: Minimum confidence for segments
  • Maintains full backward compatibility
  • Added CLI arguments for new functionality

Addresses: OpenAI Whisper Discussion #679 - Hallucinations & Repetition Loops

- Created whisper/enhancements module for enhanced functionality
- Implemented HallucinationDetector with multi-method detection:
  * Pattern-based detection (YouTube artifacts, repetitive phrases)
  * Statistical analysis (compression ratios, log probabilities)
  * Repetition analysis (looping behavior detection)
  * Temporal analysis (silence-based detection)
- Added ConfidenceScorer for comprehensive transcription quality assessment
- Enhanced transcribe() function with new parameters:
  * enhanced_hallucination_detection: Enable advanced detection
  * hallucination_detection_language: Language-specific patterns
  * strict_hallucination_filtering: Strict vs permissive filtering
  * confidence_threshold: Minimum confidence for segments
- Maintains full backward compatibility
- Added CLI arguments for new functionality

Addresses: OpenAI Whisper Discussion openai#679 - Hallucinations & Repetition Loops
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant