noScribe Changelog

version 0.5:

new WebVTT output (video subtitle format), allows also importing noScribe transcripts into EXMARaLDA
new plain text output (*.txt)
new automatic update notification on new releases
improved speaker detection: number of speakers can be defined beforehand
reduced hallucination and looping by adding a VAD filter
CUDA support now non-beta
small fixes with hebrew language setting, chinese UI locale, requirements file, etc.

Windows: beta version to test CUDA support (acceleration with NVIDIA graphics cards)

MacOS: Solves a bug where speaker-detection would become unreliable with MPS-acceleration (a switch to torchaudio 2.1.0 rectified this).
Windows: no changes

Windows: bugfix, rectifies a problem in combination with NVIDIA graphics cards
macOS: First beta release. Solved a bug with macOS Sonoma where noScribe would not react to the mouse.

much improved speaker detection/separation
new option to mark pauses (sections with no voice activity) in the transkript
new option to mark overlapping speech (experimental)
new option to include timestamps in the transcript
new noScribe Editor app to check and correct transcripts (no MS Word-Macros anymore)
noScribe now outputs an HTML-file which can be opened in every major word editor (MS Word, LibreOffice, OpenOffice...) or QDA-software package
many changes under the hood to prepare for an upcoming macOS-version and improve reliability and quality of the transcription
switched from "whisper.cpp" to "faster-whisper" as the basic framework (mainly because of the more precise timestamps)
macOS: First alpha release

new:

Translations of the user interface into Spanish, French, Italian, Japanese, Portuguese, Russian, Chinese. Thank you, mlynar-czyk, for this contribution! Be aware: These translations have been generated with a clever use of chatGPT. Please report any errors that you will find and make – if possible – a pull request with a better translation.
Added hyperlinks to the main window. You can now open the finished transcript directly by clicking on the filename in the log.
Improved speaker identification, especially in situations with quick changes (by reducing "max-len" in whisper to 30).
Installer now runs without admin rights. You should be able to install noScribe on a computer where you don’t have administrator privileges (i.e., because the machine is managed by the IT-department of your university). Thanks you, BabyFnord, for this suggestion!

fixes:

To solve the problem described in issue kaixxx#2 (transcription failing with error 3221225794 or 3221225501), I have now included a version of whisper.cpp that supports older hardware (non AVX2). NoScribe selects automatically which version to use. Be aware though that using such old hardware will result in a very slow transcription.
Corrected UTF-8 encoding error that resulted in a failing transcription in some languages (i.e., Japanese, Hungarian). Thank you to the two people reporting this problem via e-mail!
fixed: Auto save was saving too often during transcription.
fixed: Play along function in Word sometimes not finding the beginning of the transcription
fixed: Funny mistake in readme ("sensible data" instead of "sensitive data"). Thanks TheOnlyWayUp!