v6.6.1 - improve ux

Changes

• Rearrange widgets to maximize space usage.
• Move transcription status messages to transcription button to improve UX.

v6.6.0 - cancel transcriptions

1. Minor shutdown robustness.
2. Add the ability to cancel a long transcription mid-processing.

v6.5.0 - installers baby!

1. Replaced PyTorch with ctranslate2 for CUDA detection - removing the heavy PyTorch dependency from the install process
2. Migrated window/settings persistence from config files to QSettings - with proper validation and fallbacks for model, device, and quantization settings
3. Updated dependencies and simplified the installation script
4. Improved resource path handling for installed vs development environments
5. Add installers to releases.

v6.4.0 - less duct tape

Release Notes

Improvements

• More reliable model versioning: Replaced Python object ID-based model tracking with UUID versioning
• Improved thread cleanup: Recording threads now use explicit stop events and cleanup signals
• Dependency injection support: Controller now accepts optional injected dependencies

Changes

• Default quantization changed from float32 to bfloat16 for better performance on supported hardware
• Clipboard window now hidden by default on startup

New Installer

Basic transcription functionality, no processing of files, requires .net
EasyInstaller - Limited Features.zip

v6.3.0 - hotkey support

Thanks to @StopRun1 for the hotkey support. Just press F9 to start and stop recording.

• Also added drag and drop functionality for transcribing audio files.
• Refactored to save the program some consternation.

v6.2.0 - batch transcription

• When transcribing a file add a batch setting to greatly increase the speed.
• Big improvements to the GUI.

Full Changelog: v6.1.0...v6.2.0

v6.1.0 - support python 3.13

• Transcribe audio files
• Support python 3.13
• Update dependencies including but not limited to using torch 2.9.0.

v5.4.0 - robustness

Key Changes

Centralized Logging Infrastructure: Added a new logging_config.py module that sets up rotating file handlers with daily log files, replacing scattered logging.getLogger() calls throughout the codebase. All modules now use a consistent logging pattern with appropriate log levels for file output (INFO) and console output (WARNING).

Custom Exception Hierarchy: Introduced a new exceptions.py module with domain-specific exceptions (ModelLoadError, AudioRecordingError, AudioSaveError, TranscriptionError, ConfigurationError) that inherit from a base TranscriberError class. This enables more precise error handling and clearer error messages to users.

Temporary File Management: Added a singleton TempFileManager class that tracks all temporary WAV files, ensures proper cleanup on application exit via atexit, and provides thread-safe file creation and deletion. This prevents temp file leaks that could occur when transcription errors interrupted normal cleanup.

Model Metadata Refactoring: Extracted model information (names, translation support, quantization overrides) into a dedicated ModelMetadata class with a clean data-driven approach using dataclasses, removing hardcoded lists scattered across the GUI code.

Improved Error Propagation: The controller now emits structured error_occurred signals that the main window displays via QMessageBox dialogs, giving users clear feedback when operations fail. Status bar messages are now user-friendly summaries while full details go to logs.

Resource Cleanup Hardening: Added timeout-based thread termination with fallback to terminate() if threads don't stop gracefully, and improved mutex handling with proper try/finally patterns to prevent deadlocks.

v6_0_0 - transcribe audio files

Revise README for Faster Whisper Transcriber

Updated README to reflect new features, requirements, and installation instructions.

v5.3.1 - Add Whisper Turbo

Added the Whisper Large v3 Turbo model. It's faster than the non-turbo model but about the same as the "distilled" v3 model. That said, the distilled model and turbo model are very similar.

• Whisper Turbo can't translate so this was addressed 5.3.1 versus 5.3.