02 / Personal tool
LocalTranscriber
A local workflow for turning long lecture and meeting recordings into usable text. I defined the requirements, used AI assistance to develop and troubleshoot the workflow, and tested it on my Windows computer.
- Status
- Working tool · ongoing improvements
- Platform
- Windows / Python
- Focus
- Local transcription
Workflow
The decisions behind a local tool.
The workflow combines Whisper-based transcription, pyannote speaker processing, FFmpeg, audio chunking, and logging. Local processing is central to the tool, while visible progress helps make long recordings easier to work with.
- 01Prepare and chunk audio with FFmpeg
- 02Transcribe using a Whisper-based workflow
- 03Run speaker-processing stages and track progress in logs
- 04Produce separate raw and cleaned Markdown outputs for review
Results & lessons
Useful text still needs a careful read.
Recent processing logs and output files confirm transcription, speaker-processing stages, and separate raw and cleaned Markdown outputs. Transcripts still need manual review. Accuracy percentages remain unpublished until benchmark evidence is checked.
Audio quality, language, speaker separation, reliable dependencies, and visible processing progress all affect usability. Those are the areas I continue improving.
Next project
Production and logistics optimisation
A university assignment on resources and delivery requirements.
View production and logistics