On-Device Speech-to-Text Transcription

How does on-device transcription work in Passcut?

Passcut transcribes speech on-device using Apple SpeechAnalyzer on macOS 26. You pick the language spoken in the clip. Extra clips transcribe only when you ask. Media is not uploaded to a Passcut cloud.

Running natively on Apple Silicon, SpeechAnalyzer processes audio tracks directly from imported video containers via AVFoundation, producing precise word-level timing data for text-based video editing. Apple’s speech model uses one language per media file.

How is the transcript linked to the video timeline?

Passcut establishes bidirectional sync between the generated transcript and the assembly timeline, allowing you to seek video frames by clicking transcript text and delete video segments by rejecting transcript lines.

When you strike out a sentence or paragraph in the transcript editor, Passcut immediately removes those video frames from the primary assembly track while keeping connected B-roll, titles, and audio tracks aligned.

macOS 26 · Apple Silicon

Download Passcut for Mac

Native macOS. Apple Silicon. Transcript-first rough cuts, then hand off to Final Cut Pro, DaVinci Resolve, or Premiere Pro.

Download Passcut Buy

Requires macOS 26 or later and Apple Silicon. Footage does not go to a Passcut cloud.Compare plans.

#Frequently Asked Questions

What framework powers Passcut transcription?

Passcut uses Apple SpeechAnalyzer, a native macOS framework optimized for Apple Silicon Neural Engine hardware.

Does transcription require an internet connection or API key?

Transcription itself runs on-device with Apple SpeechAnalyzer. First transcription may download Apple speech assets. No third-party speech API key is required.

How is the transcript linked to video playback?

Clicking any word or line in the transcript instantly seeks the video playhead to that exact frame. Rejecting a transcript line removes the corresponding video segment from the cut.

What language can I transcribe?

You pick it. Ask every time (default), reuse the last language, or always use one. One language per media file — Apple’s speech model does not switch mid-file.