Releasing version v0.3.12

• Switch transcription to Whisper models hosted on Hugging Face Hub
• Use VAD-based audio segmentation prior to transcription and fix device placement
• Extract word timestamps via attention-based DTW alignment with fallbacks (experimental)
• Improves reproducibility and avoids build isolation issues in CI
• Restore TF Metal device support and clean up dependencies

Releasing version v0.3.11

Minor release for reinstating the previous stretching mechanism

Releasing version v0.3.10

• Support py312 and Tensorflow 2.19
• Upgrade OpenAI Whisper
• Update base images for Ubuntu

Releasing version v0.3.9

• Support Tensorflow 2.15
• Add the option to use the FB M2M100 model for translation
• Introduce the ability to use adjacent cues as prompts during transcription
• Include optional word timecodes in forced alignment and transcription outputs

Releasing version v0.3.8

• Enhanced subtitle generation with improved time codes
• Introduced a chunking option to limit segments by maximum character length
• Included an option to use original subtitles as prompts when transcribing individual audio segments
• Included an option to set a global prompt for transcription across the entire audio
• Added support for Whisper's turbo model for faster and more efficient transcriptions

Releasing version v0.3.7

• Expose configurations for various processing timeouts on API/CLI
• Enhance encoding for intermediate subtitle files generated during forced-alignment
• Modernise direct dependencies and GH action dependencies
• Update the Helm chart to use Deployment

Releasing version v0.3.6

• Support py311
• Update dependencies

Releasing version v0.3.5

• minor release for multi-platform images

Releasing version v0.3.4

• Upgrade Whisper and show the transcription progress bar in debug mode
• Update base images for Ubuntu, Debian and Fedora Docker images
• Deprecate egg distribution and modernise required dependencies

Releasing version v0.3.3

• Fix dependency misplacement and update project metadata