Skip to content

Data and Output Locations

Preferences

  • ~/Library/Preferences/com.openscreech.app.plist

ASR Models

  • ~/Library/Application Support/FluidAudio/Models/parakeet-tdt-0.6b-v3-coreml

Optional Diarisation Models

  • Base directory: ~/Library/Application Support/FluidAudio/Models/
  • Models are downloaded only after selecting a diariser and clicking its Download button in Recording Settings → Speaker Diarisation Provider.
  • The selected diariser's exact destination is shown beneath its provider controls and can be revealed with Show in Finder, including before it is downloaded.

Optional Screechback Speech Models and Voices

  • Base directory on macOS: ~/.cache/fluidaudio/Models/
  • Kokoro ANE resources are stored beneath kokoro-82m-coreml/ANE or kokoro-82m-coreml/ANE-zh for the selected English or Mandarin model.
  • PocketTTS resources are stored beneath pocket-tts/v2.1, with the selected language/model variant below it.
  • Supertonic-3 resources are stored beneath supertonic-3.
  • Models are downloaded only after selecting one and clicking its download button in App Settings → Local Speech Models & Voices or in a profile's Screechback section.
  • The UI shows every remote FluidInference source and the exact local destination. Installed resources can be revealed with Show in Finder or removed with the confirmed Delete Model & Voices… action.
  • Kokoro downloads the complete published voice catalogue for the selected model. PocketTTS and Supertonic-3 downloads include their available voices or voice styles.

Model deletion is scoped to the selected model's owned directory and includes that model's voices. It preserves sibling variants and shared parent resources. OpenScreech unloads the selected model before deletion and updates its status immediately; it can be downloaded again normally.

Screechback generates WAVs in a dedicated folder inside macOS's per-user temporary directory, typically /var/folders/.../T/OpenScreech/Screechback/, using openscreech-screechback-*.wav and openscreech-screechback-test-*.wav names. Normal playback removes its file when playback ends or stops; the tester removes its replay file when replaced or when the test window closes. App Settings → Local Speech Models & Voices → Temporary Screechback Audio shows the exact directory, count, and size and provides manual Finder and deletion controls. Generated speech is not retained in Saved Outputs or Notes.

FluidAudio model components are cached under ~/.cache/fluidaudio/Models/. Alongside the compiled components each model loads, that cache can hold uncompiled .mlpackage sources and components superseded by a FluidAudio upgrade. App Settings → Local Speech Models & Voices → Unused Model Components reports their total size and offers Finder access and a confirmed deletion that leaves installed models and voices intact.

Long-Form Recording Output

Default root:

  • ~/Documents/OpenScreech Recordings

Configure this location under Recording Settings → Storage.

Per-session folder example:

  • Recording-20260413-153000

Optional diarisation artifact:

  • transcript.speakers.txt

Multitrack sessions

Multitrack recording uses the same recording root. Combined outputs are placed at the session root; individual audio and exports are under tracks/01/, tracks/02/, and so on. Each enabled track contains a continuous lossless recording.caf. Folder numbers identify tracks independently of editable labels, list order, and input assignments.

session.json records the interface and track mapping, capture details, and errors. A hidden .chunks/ folder holds temporary audio queued for recognition and intermediate recognition data. It is removed after successful finalisation and retained when errors need investigation.

The persistent track list and selected interface are stored in app preferences, separately from the audio files. Track names and input overrides survive interface changes. See saved channel names and disconnected interfaces for reconnect behaviour and current limitations.

Existing File Transcription Output

Default root:

  • ~/Documents/OpenScreech File Transcriptions

Dictation Saved Outputs (Per Profile, if enabled)

Default root:

  • ~/Documents/OpenScreech Captures

Artifacts can be split into configurable subdirectories (audio, original transcript, word-replaced transcript, LLM transcript).

Notes (Per Profile, if enabled)

Default root (changeable in the Notes window, e.g. to a synced iCloud Drive folder):

  • ~/Documents/OpenScreech Notes

One folder per capture inside a category folder (default Inbox), named yyyy-MM-dd-HHmmss-<6-char ID> and holding metadata.json (timestamps, tags, profile), transcript.txt, transcript.replaced.txt, transcript.llm.txt, and capture audio without conversion (audio.wav for dictation or the original extension for server uploads). A tags.json registry lives at the root.

See Notes for the complete bundle model and Notes window behavior.

HTTP Request Capture (if enabled)

Configurable root directory in HTTP Transcription Service.

Quick Tips

  • The preferences plist contains configuration, not your recordings, Notes, saved transcripts, or downloaded FluidAudio models. A complete backup must include the output and model roots you actually use.
  • Changing an output location affects future writes. Existing recording sessions, file transcriptions, captures, and Notes are not moved automatically.

Visual Notes

  • TODO: add screenshot of Finder tree for one complete workflow.