Profanity and Detection

Profanity & Detection

Censorarr combines Whisper transcription, profanity matching, rescue passes, and optional subtitle evidence.

Profanity dictionary

The default configured profanity file is:

/config/en.json

Default minimum severity:

profanity:
  min_severity: 3

Lowering the threshold can make filtering more aggressive.

Matching

Censorarr supports:

• direct dictionary matches
• word-family aliases
• fuzzy matching
• multi-word windows
• rescue passes around uncertain regions

Example:

profanity:
  max_word_window: 4

Rescue mode

Default example:

rescue:
  enabled: true
  confidence_trigger: 0.18
  fuzzy_confidence_ceiling: 0.70
  fuzzy_similarity: 0.72
  prefer_center_channel: true

The rescue prompt intentionally asks Whisper to transcribe profanity verbatim rather than euphemize it.

Precision mute alignment

Enabled by default in the example configuration:

precision_alignment:
  enabled: true
  padding_before_ms: 25
  padding_after_ms: 40

It tightens mute timing by protecting neighboring words and looking for nearby low-energy waveform boundaries.

If disabled, Censorarr falls back to the broader legacy padding:

padding_before_ms: 120
padding_after_ms: 160

Subtitle evidence

Subtitle Assist can clarify uncertain dialogue and shifted timing, but Whisper remains part of the detection pipeline.

Custom profanity and exceptions

Runtime customization files are stored under /config, including:

/config/custom_profanity.json
/config/profanity_overrides.json
/config/user_exceptions.json

Keep /config persistent across updates.

Accuracy expectations

No ASR system can guarantee perfect detection. Use Dry Run and Review Mode on representative media before applying settings across an entire library.

Dictionary interface

The Dictionary page lets you review built-in and custom detection patterns, severity levels, detection scope, and permanent exceptions.