Queue the work
Choose multiple audio files or drag them into the app. Jobs run one at a time and can be removed while queued, running, or complete.
Drop a supported audio file into AFK, let the on-device speech engine work through it in chunks, and save the result as text, subtitles, or timestamped JSON.
Interviews, lectures, voice memos, and other long audio can be queued from the Transcribe tab. AFK streams source audio in 120-second chunks, transcribes each chunk, and shifts word timings back onto the full recording timeline.
Choose multiple audio files or drag them into the app. Jobs run one at a time and can be removed while queued, running, or complete.
When the model returns word timings, AFK can show timestamped segments and use them for SRT or JSON export.
Save TXT, SRT, or JSON, or pick a folder and AFK saves each finished file there. Unsaved non-empty results stay in the local cache for one hour; saving adds a lightweight reference to the Files tab in Notes.
The drop target explicitly recognizes these extensions, and the file picker accepts audio types recognized by macOS.
.m4aCommon voice-memo and compressed audio files..wavUncompressed PCM audio..mp3Compressed audio..aiff / .aifAIFF audio files..cafCore Audio Format files..flacLossless audio files.The app and the CLI also take video with an audio track: .mp4, .mov, and .m4v.
The default Parakeet v3 model is multilingual across 25 European languages. AFK can also use an English-only Parakeet v2 model from the app's settings.
The first run needs the network to download and optimize the model. After that, file reading and speech recognition run locally. License activation and Sparkle update checks are separate network operations.
M4A, WAV, MP3, AIFF, CAF, and FLAC. It also takes MP4, MOV, and M4V video that has an audio track.
AFK has no set limit. It reads a file in chunks of about two minutes, so a three-hour recording takes longer but works the same way as a short one.
No. The speech model runs on your Mac and the recording is not uploaded. AFK uses the internet to download the model during setup, activate your license, and check for updates. The privacy page has the details.
AFK skips that part, keeps going, and marks the transcript as incomplete. The JSON export lists the time range that was skipped, so you know where to listen again.
Yes. Add as many as you like and AFK works through them one at a time. Choose an output folder first and each finished transcript is saved there. An existing file is never overwritten.
English and 24 other European languages with the default model. The dictation guide lists them.
Install AFK, set up the model once, and make the next long audio file a local job.