Transcribe Tool | Portfolio

The Mission

Voice messages are great for capturing thoughts, but terrible for searching them later. I built this pipeline to automatically process my recordings, transcribe them using Whisper, and translate/summarize them using a local LLM. It turns audio chaos into structured XML data.

See it in Action

Live execution on local machine:

bello@bellos-sys — python3

(.venv) bello@bellos-sys:~/code$ python3 src/main.py --model large-v3-turbo

Found 1 new audio file(s) to transcribe:
  - test-voice.m4a
----------------------------------------

Processing: test-voice.m4a
clean_audio() took 0.50 s
process() took 3.88 s
Transcription: "Okay, das ist jetzt hier nochmal eine Test-Memo..."
Saved to output/output.xml

========================================
Transcription Summary:
Total files: 1 | Successful: 1
========================================

Translating 1 entries...
generate() took 2.72 s

(.venv) bello@bellos-sys:~/code$ cat output/output.xml
<entry id="ID-9a65ce58">
  <transcription lang="de">
    Okay, das ist jetzt hier nochmal eine Test-Memo...
  </transcription>
  <translation>
    Okay, this is just another test memo...
  </translation>
</entry>

View Source on GitHub