Synthalingua 1.2.1: New Backends, Portability, and Major Fixes!

Hello everyone! This is a massive update for Synthalingua that combines several major releases into one. These changes introduce powerful new features, dramatically improve performance and hardware support, and make the initial setup process smoother than ever. Read on for the full details!

New Features & Major Enhancements

• Multiple AI Backends (--model_source): You can now choose from three different transcription engines to get the best performance on your specific hardware!
  • whisper: The original, reliable OpenAI model. A great baseline for compatibility.
  • faster-whisper: The new recommended default! This is a complete re-implementation that is up to 4 times faster and uses half the VRAM with the same accuracy. Perfect for NVIDIA users or anyone needing a speed boost.
  • openvino: A game-changer for Intel users! This backend is optimized for Intel hardware, unlocking the full potential of your device.
• Intel Hardware Acceleration: Thanks to the new OpenVINO backend, Synthalingua now officially supports and is optimized for a wide range of Intel hardware. Use the --device flag with new options like intel-igpu (integrated GPUs), intel-dgpu (discrete GPUs like Intel Arc), and intel-npu (the new Neural Processing Units in Core Ultra processors for super-efficient AI).
• Model Quantization (--compute_type): You can now run the larger, more accurate models even on hardware with limited VRAM! Quantization reduces the model's memory footprint and increases speed, making high-quality transcription more accessible to everyone.
• Intelligent Mode for Subtitles (--intelligent_mode): When generating subtitles with --makecaptions, this new mode will automatically detect if the transcription quality is low or if the AI is repeating itself. If it is, Synthalingua will automatically retry that specific audio segment with a larger, more powerful model until the result is perfect. This ensures the best possible accuracy without any manual effort.

Core Engine Improvements & Bug Fixes

• Major Memory Leak Fixed: We've fixed a critical stability issue where the application would not release VRAM/RAM after a transcription task was finished. The subtitle generation process now runs in an isolated process, which guarantees that all memory is freed after each task. This prevents crashes, especially when generating subtitles for long files.
• Unified Backend Logic: To support the new multi-backend system, the core transcription engine has been re-engineered to seamlessly handle the different APIs of Whisper, FasterWhisper, and OpenVINO, ensuring consistent and reliable results no matter which engine you choose.
• Accurate FasterWhisper Language Detection: The audio processing pipeline for the new FasterWhisper backend has been carefully tuned to ensure its language detection is accurate and reliable right out of the box.
• Robust Stream Transcription: The app can now handle corrupted or empty audio segments from HLS streams without crashing. It will simply print a warning and skip the faulty segment, making live stream translations more stable.
• Browser Cookie Fix: Fixed an issue where cookies extracted from a browser were not being correctly used during the interactive stream selection process.
• Improved Blocklist Filtering: The blocklist is now cached in memory and only re-read if you've modified the file, making the filtering process more efficient.

Setup & User Experience

• More Reliable Portable Version: We've completely overhauled how the portable (.exe) version is built. This ensures all necessary files are included, making the standalone application much more stable and reliable.
• New Comprehensive Source Setup Script (SourceSetUp.py): For users running from source, a powerful new setup script has been introduced. It can automatically download and install the correct version of Python in an isolated way (so it won't mess with your other projects!), create the virtual environment, install all dependencies, and even download FFmpeg and yt-dlp for you.
• Greatly Enhanced Command-Line Help (--help): We've rewritten the help descriptions for almost all commands to be more detailed and user-friendly, with clear examples and recommendations to help you get the best results.
• Streamlined Intelligent Mode: The "Intelligent Mode" is now controlled exclusively via the --intelligent_mode command-line flag, making it easier to use in automated scripts.
• Robust Vocal Isolation (Demucs) Path Handling: A new helper has been added to reliably find the Python environment for Demucs, making the --isolate_vocals feature much more stable and easier to use.
• Automatic Model File Migration: To keep things organized, the app will now automatically move your existing AI model files into the new, cleaner models/Whisper/ subdirectory. This is a one-time process to help existing users adapt to the new structure without any hassle.

Thank you for your continued support and feedback. These updates have made Synthalingua more powerful, stable, and accessible than ever before!