v1.4.1

Bugfix Release!

Fixes...

• speaker recognition running on CPU when it should run on GPU. Thanks @Biga73 for reporting #109
• increase in repetitions in transcripts (restores default temperature behavior). Thanks @KlausFarmer for reporting #111
• transcription folder button not working in some linux distros. Thanks @Dorianux for reporting #113

v1.4.0

This new major release of aTrain features a complete rewrite of the user interface in NiceGUI, which greatly simplified the codebase.

This also fixed issues with GPU-transcription not producing output #84

In addition, this new version includes:

• Improved speaker diarization using the newest pyannote.audio 4.0.3

• Support for Nvidia RTX 50xx series

v1.3.1

Bugfixes only in this release.

This release fixes some bugs regarding error handling in child processes and updates the torch dependency to address recent security concerns.

v1.3.0

Major Bugfixes and Development Guidelines

Those were some rough months. Updating the backend to the newest upstream software and faster-whisper models as well as our rewrite of the codebase, seperating frontend GUI and backend in atrain_core, together intrudeced similar looking bugs that each only appeared with some of the files.

Thanks to the community feedback and provision of problematic files, we found our own backend bug and then, later, also the upstream bug in faster whisper.

This should now run smoothly.

We also introduced guidelines on how we will do branching in the future and how to contribute.

Thanks for providing feedback, logs, and problematic files for testing:
@Beefthong @Biga73 @OhsawaSenpai @allan-simon @architeck @aereimer @happypig @adlihm @OhsawaSenpai @wenyuan-wu @MonikaBarget @fstolba

v1.2.1

Release v.1.2.1

We are happy to announce the release of our newest improvements. This version includes some additional features and bug fixes, for example:

• Replacement of multiprocessing with stoppable threads
• Possibility to add an initial prompt in advanced settings thanks to the contribution of @Bytestorm5
• Add float32 to possible inputs
• Add note for macOS users that file names with white spaces can lead to errors
• Fix model hash error
• Update FAQs

v1.2.0

Release v1.2.0: Engine Upgrade

With this release, we also welcome @andreafooo to our team who did a lot of the heavy lifting for it; glad to have you! What also makes us really happy is the increased involvement of all the contributors for the growing aTrain user community.

Also, as a shameless plug, it is really cool that aTrain is now recommended at Harvard University (and many other Universities worldwide) for the local transcription of sensitive audio material

But on to the release, and it is a big one, as it includes a completely rewritten backend, lots of new features and improvements and the use of faster-large-v3-turbo as the default model.

Major Improvements

• Completely new backend aTrain_core
• Support for faster-whisper large-v3
• New default model for a great balance between speed and accuracy: faster-whisper large-v3-turbo
• Support for distil-whisper models single language models for large speed improvement. Currently only for English and marked as beta as there is an illusive bug that leads to sometimes not showing the completion in the GUI, while the transcript is already finished and available in the folder.
• Introduction of our Model Manager: Download only models you really use, reducing the installer size greatly (we include speaker diarization model and faster-large-v3-turbo as the defaults)
• Halfed installer size thanks to model manager
• Rewrote transcription time estimate function, now estimated live and it should be very accurate

Minor improvements

• Updated faster-whisper to 1.0.2
• Updated pyannote-audio to 3.2.0

Major bugfixes

• aTrain no longer crashes when having special characters in the filename. Thanks to @wenyuan-wu and @hirowa for figuring this out and to @SjDayg for fixing it
• Numpy error when installing fixed, thanks @samfisherirl

v1.1.0

This version of aTrain introduces two new features:

• Advanced settings for specifying the compute device (CPU vs. GPU) and the compute type (int8 vs. float16) used for transcription.
• Additional output format: SRT files for subtitles

The code base of this release is also restructured to a proper Python project installable with pip.

v1.0.0

Initial release