FFFF echofield/docs/SETUP.md at main · shroominic/echofield · GitHub
[go: up one dir, main page]

Skip to content

Latest commit

 

History

History
88 lines (53 loc) · 2.54 KB

SETUP.md

File metadata and controls

88 lines (53 loc) · 2.54 KB

Setup

Release Build

  1. Download EchoField.dmg from the latest GitHub release.
  2. Drag EchoField to Applications.
  3. Control-click the app and choose Open the first time if macOS identifies it as an unsigned community build.
  4. Complete the three Setup Assistant tabs.

The runtime installer adds uv and FFmpeg through Homebrew and creates a Python environment under:

~/Library/Application Support/EchoField/Runtime/

The Qwen model downloads on the first transcription and is cached by Hugging Face.

Recorder

The default profile expects:

/Volumes/H8/RECORD/*.WAV

Set Volume name and Recording folder to match the names visible in Finder. EchoField ignores AppleDouble ._* files.

Diary Output

Choose any existing folder. EchoField creates a child folder containing:

Daily/YYYY/YYYY-MM-DD.md
Recordings/YYYY/YYYY-MM-DD/<recording>.md
Transcripts/YYYY/YYYY-MM-DD/<recording>.md

If the chosen folder is an Obsidian vault, note links open in Obsidian. Otherwise they open in the default Markdown editor.

Optional Speaker Separation

The runtime installer includes pyannote, but its models require upstream access approval and a Hugging Face login:

hf auth login

Source-checkout contributors can install the dependency directly with uv sync --extra diarization.

Enable Speaker separation after the token is available. Speaker labels are anonymous (SPEAKER_00, etc.) unless a future identity workflow maps them explicitly.

Optional Cleanup

Install and start Ollama, then fetch the default model:

ollama pull qwen3:1.7b

Enable LLM transcript cleanup. Raw and diarized transcripts remain unchanged.

Storage

Fresh installs use:

~/Library/Application Support/EchoField/

Upgrades from H8 Diary continue using the existing directory to avoid duplicating large archives.

Troubleshooting

The app waits for a recorder

Confirm the Finder volume and recording-folder names match Settings. Disconnect and reconnect after changing them.

Cleanup says Ollama is unavailable

Open Ollama and run ollama list. Press Run Now to retry missing stages.

Diarization fails

Confirm hf auth whoami succeeds and that access to the configured pyannote model has been accepted upstream.

A file stays on the recorder

This is intentional when any archive, transcript, diary, checksum, or source-stability check fails. Open Activity, fix the failed stage, and run again. Do not manually delete the recorder copy until the local WAV and diary output are verified.

0