kaixxx/noScribe

📋 What's Changed

• True headless mode for servers
• Ffmpg to av, thanks @mutlusun. This also removes Rosetta2 requirement for macOS.
• some small issues fixed
• Full Changelog: https://github.com/kaixxx/noScribe/compare/v0.7...v0.7.2
• Download and install: https://noscribe.de/

📋 Changes

• New batch transcription: process several files at once
• Improved speaker identification
• Improved handling of punctuation issues
• Command line interface for scripting
• many small fixes, improvements and updates

📋 Changes

• 3x faster transcription
• New: Install custom whisper models
• New option to include or exclude disfluencies and filler words
• Editor: Search & Replace function to quickly change speaker names etc.
• Many small fixes, improvements and updates
• Installation instructions: https://github.com/kaixxx/noScribe/tree/main#download-and-installation

📋 Changes

• WebVTT output, allows importing noScribe transcripts into [EXMARaLDA](https://exmaralda.org)
• plain text output (*.txt)
• improved speaker detection: number of speakers can be defined beforehand
• reduced hallucination and looping by adding a VAD filter
• automatic update notification on new releases
• editor opens automatically (can be disabled in advanced options)
• CUDA support now non-beta
• small fixes with hebrew language setting, chinese UI locale, Linux support, etc.

📋 Changes

• much improved speaker detection/separation
• new option to mark pauses (sections with no voice activity) in the transcript
• 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)

📋 Changes

• Translations of the user interface into Spanish, French, Italian, Japanese, Portuguese, Russian, Chinese. Thank you, [mlynar-czyk]( https://github.com/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](https://github.com/BabyFnord), for this suggestion!
• To solve the problem described in issue https://github.com/kaixxx/noScribe/issues/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
• + 1 more