Transcribe the long recording in chunks.

AFK reads supported audio, or the audio track of a video, as a stream of 120-second chunks, transcribes one file at a time, and rebuilds a continuous timeline for results with word timings.

Start a transcription

  1. Open AFK's Transcribe section and wait for the speech model to be ready.
  2. Drag supported audio or video files into the drop zone, or choose Choose files… to select multiple files.
  3. AFK adds each file to the queue and works through the jobs one at a time.
  4. When a result is ready, copy it, show timestamped segments when timings exist, or open the Save menu.

The file flow reads audio through macOS's audio framework, takes the audio track out of a video through Apple's media framework, and converts it to 16 kHz mono samples for the speech engine.

Supported inputs

The app and the CLI accept these audio and video extensions:

ExtensionInput
.m4aMPEG-4 audio files such as many voice memos.
.wavWaveform audio.
.mp3MP3 audio.
.aiff / .aifAIFF audio.
.cafCore Audio Format.
.flacFLAC audio.
.mp4MPEG-4 video with an audio track.
.movQuickTime movie with an audio track.
.m4vM4V video with an audio track.

Choose files… offers audio files plus MPEG-4, QuickTime, and M4V movies. For a video, AFK extracts the audio track into a private temporary folder and removes it when the job succeeds or fails. A video without an audio track shows a clear error, and other video containers such as MKV, WebM, or AVI are not supported. The CLI accepts the same files; see the CLI input boundary.

How long recordings are handled

AFK's audio reader streams the source rather than allocating one giant buffer. Each chunk is approximately 120 seconds of source audio. After a chunk finishes, the queue advances and progress is reported by source time. Word timings from each chunk are shifted by that chunk's position so the result has one global timeline.

The queue is serial: one file runs at a time. Remove a pending, running, or finished job with the × control. Removing a running job cancels that job without cancelling the rest of the queue; cancellation can wait for the current chunk to finish.

Save formats

FormatWhat is saved
.txtThe transcript text as UTF-8 plain text.
.srtSubRip subtitle segments generated from word timings. The Save menu shows this when the result has timings.
.jsonA pretty-printed transcript object with text, duration, and word-level timing records.

When you save, AFK writes the file you choose and adds a lightweight reference to the Files tab in Notes. The reference records the source filename, saved path and format, source duration, processing time, and word count. It does not become the permanent home for the transcript text.

Save finished files to a folder

To skip the Save menu, choose Save completed files to a folder… in Transcribe, pick a folder, and choose TXT, SRT, or JSON. Each job you queue from then on is written to that folder in that format as soon as it finishes, and gets a reference in Notes like a manual save. Use Change… to pick another folder, or Ask each time to go back to saving by hand.

AFK never overwrites an existing file: if the name is taken, it saves name-2. If SRT is chosen for a result without timings, or the file cannot be written, the job shows an error and its Save menu stays available.

Unsaved results and history

A non-empty result that you have not saved is cached locally for one hour after completion. The result card shows a countdown and says when it is saved. Save it before the window expires if you want a durable copy. Empty results are session-only.

AFK restores fresh cached results after a relaunch, and removes expired or unreadable cache files. Settings → Data → Delete history and file cache removes cached results and the references in Notes; files you already saved remain where you put them.

Supported languages

Parakeet v3 is described in the source as multilingual across 25 European languages:

  • English
  • Bulgarian
  • Croatian
  • Czech
  • Danish
  • Dutch
  • Estonian
  • Finnish
  • French
  • German
  • Greek
  • Hungarian
  • Italian
  • Latvian
  • Lithuanian
  • Maltese
  • Polish
  • Portuguese
  • Romanian
  • Russian
  • Slovak
  • Slovenian
  • Spanish
  • Swedish
  • Ukrainian

The app can steer v3 toward a selected language's alphabet. The CLI can accept an ISO 639-1 language hint with --lang; see the CLI guide.

Offline boundary: after the model is downloaded and ready, this transcription work runs locally. The first setup, license activation, and Sparkle update checks are the network-dependent parts.