Skip to content

Artifacts & Caching

Every edit or transcribe run saves artifacts to ~/.praisonai/editor/{filename}/.

What is saved

~/.praisonai/editor/{filename}/
  ├── probe.json           ← media metadata
  ├── transcript.json      ← word-level timestamps (CACHED)
  ├── transcript.srt       ← SRT subtitles
  ├── transcript.txt       ← plain text
  ├── plan.json            ← edit decisions (keep/remove segments)
  └── content_blocks.json  ← detection blocks (only with --detector)

~/.praisonai/editor/.demix_cache/{sha256}/
  ├── vocals.wav           ← separated vocals (CACHED)
  └── no_vocals.wav        ← instruments (CACHED)

Caching behaviour

sequenceDiagram
    participant CLI
    participant Cache
    participant API

    CLI->>Cache: Is transcript.json present?
    alt Cache hit
        Cache-->>CLI: ↻ Load transcript
    else Cache miss
        CLI->>API: Call Whisper API
        API-->>CLI: Transcript
        CLI->>Cache: Save transcript.json
    end

    CLI->>Cache: Is .demix_cache/{hash} present?
    alt Stem cache hit
        Cache-->>CLI: ↻ Reuse vocals.wav + no_vocals.wav
    else Cache miss
        CLI->>CLI: Run Demucs (~10 min)
        CLI->>Cache: Save stems
    end

Disable artifact saving

praisonai-editor edit podcast.mp3 --no-artifacts

Clear caches

# Remove transcript cache for a specific file
rm ~/.praisonai/editor/podcast/transcript.json

# Remove all stem caches
rm -rf ~/.praisonai/editor/.demix_cache/

# Remove everything
rm -rf ~/.praisonai/editor/

probe.json example

{
  "path": "podcast.mp3",
  "duration": 1823.45,
  "has_video": false,
  "audio_codec": "mp3",
  "audio_sample_rate": 44100,
  "audio_channels": 2,
  "size_bytes": 44352512
}

plan.json example

{
  "original_duration": 1823.4,
  "edited_duration": 1680.2,
  "removed_duration": 143.2,
  "removal_summary": {"filler": 12.3, "silence": 122.8, "repetition": 8.1},
  "segments": [
    {"start": 0, "end": 5.2, "action": "keep", "category": "content"},
    {"start": 5.2, "end": 5.8, "action": "remove", "category": "filler", "reason": "Filler: 'um'"}
  ]
}

content_blocks.json example

{
  "resolved": [
    {
      "start": 0.0, "end": 367.0,
      "type": "singing",
      "detector": "ensemble",
      "rms_db": -18.5,
      "confidence": 0.91
    }
  ]
}