VidEdit Jetpack Compose (Android)

Core Android port of your text-based video editor:

• Video selection from storage
• Audio extraction with ffmpeg
• Transcription with whisper.cpp CLI + model
• Word-based cut marks (delete words -> delete time segments)
• Export edited MP4 with ffmpeg

Current scope

This is the core pipeline implementation (not full polished editor UI):

1. Pick video
2. Transcribe
3. Tap words to mark cuts
4. Undo/Clear cuts
5. Export edited video

Project path

VidEditJetpackCompose/

Binary/model requirements

The app uses FFmpeg via Android SDK (FFmpegKit), so no standalone ffmpeg binary is required. Whisper CLI is packaged as native libs in the APK:

• app/src/main/jniLibs/arm64-v8a/libwhisper_cli.so
• app/src/main/jniLibs/armeabi-v7a/libwhisper_cli.so

The app expects this model in app internal storage:

• files/models/ggml-tiny-q5_1.bin

At startup it copies model from bundled assets:

• app/src/main/assets/models/ggml-tiny-q5_1.bin

If not found, the app launches but transcription/export will be blocked with a setup error message.

Important:

• libwhisper_cli.so must be an Android Linux executable for each supported device ABI.
• After adding/replacing model in assets, reinstall the app so startup extraction runs again.

Bundled now:

• app/src/main/jniLibs/arm64-v8a/libwhisper_cli.so
• app/src/main/jniLibs/armeabi-v7a/libwhisper_cli.so
• app/src/main/assets/models/ggml-tiny-q5_1.bin is included in this project.

Notes

• whisper.cpp command uses JSON output (-oj) and parses word/segment/token formats.
• ffmpeg export uses trim+concat filter graph from kept segments.
• Export output goes to app external files directory with timestamped filename.