(Linux) No punctuation or capitalization in German transcripts

[oldnick1818]

Using the ported version on Ubuntu 25.04 and the settings exported from the config below, the transcription is "raw" - containing no punctuation or capitalization.

As in German nouns get capitalized, manually correcting this is very work intensive. It would be great if the transcription could do this automatically, which - if I understand correctly - Whisper should be capable to do.

Is there a way to change this behavior?

THANK YOU!!

my configs:
(as I thought this might be related to the localization of the GUI, I tried to change it to German in the config; the GUI is still English though)

app_version: 0.6.2
auto_edit_transcript: 'True'
auto_save: 'True'
check_for_update: 'True'
last_disfluencies: 0
last_filetype: html
last_language: German
last_overlapping: 0
last_pause: 3sec+
last_speaker: '2'
last_timestamps: 1
last_whisper_model: precise
locale: DE
pause_seconds_marker: .
pyannote_xpu: cpu
threads: 16
timestamp_color: '#78909C'
timestamp_interval: 60000
voice_activity_detection_threshold: '0.5'
whisper_beam_size: 1
whisper_compute_type: default
whisper_temperature: 0.0
whisper_xpu: cpu

[mutlusun]

Running whisper also on linux and using a German audio file, punctuation and capitalization works for me very well. Can you test downloading this file and transcribing it. Is punctuation and capitalization still missing?

If it occurs only with your file, it might help to transcribe only portions of the file (set start/stop time).

[kaixxx]

Partially loosing capitalization and punctuation can happen with long transcripts. This is an inherent problem of the whisper model and not related to Linux, as far as I know.

What you can do:

• Check the "disfluencies" box ("Füllworte"). Internally, this will send a special "prompt" to the AI model which acts as an example how the transcript should look like. Since this example also contains capitalization and punctuation, it might improve your results. However, you might get some extra "ehms" in your transcript...
• Transcribe in portions as @mutlusun suggested might help indeed.
• Wait for the new version that will be released in a few days (not sure when a binary for Linux will be available), or try it directly from source: https://github.com/kaixxx/noScribe/tree/pyannote4 This contains a few updates which try to counteract the problem you've described. However, I cannot promise that it will be completely gone.

[oldnick1818]

Thank you for your fast help!

• as my audio file was only 5 minutes long, I don't think this was the problem.
• I tried the file @mutlusun supplied with the same settings and it gave correct punctuation and capitalization.
• activating "disfluencies" finally did the trick on my audio - it now has punctuation and capitalization!

I have a working solution for now and will test again without the disfluencies when the new binaries are available. Thanks again!