v25.12.33

[ROBERT-MCDOWELL]

No description provided.

[Copilot]

Pull request overview

This pull request (v25.12.33) introduces significant refactoring to the TTS (Text-to-Speech) engine architecture and audio processing pipeline. The changes focus on improving GPU/device handling, VTT subtitle generation, and text processing logic.

Changes:

• Refactored GPU policy handling to support multiple device types (CUDA, ROCm, MPS, XPU) with improved AMP dtype selection
• Moved VTT subtitle generation from inline processing to a deferred batch operation using audio file analysis
• Extracted common TTS engine methods (_set_voice, _convert_sml) to a shared utility base class
• Added text merging logic for very short rows in join_ideogramms function
• Fixed template parameter name in time format string substitution

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 19 comments.

Show a summary per file

File	Description
lib/gradio.py	Refactored event handler to chain voice list updates; simplified return values for change_gr_fine_tuned_list
lib/core.py	Added text merging pass, fixed time format parameter, refactored convert_chapters2audio to track sentences separately, extracted get_audio_duration function
lib/classes/tts_manager.py	Added create_sentences2vtt method to support deferred VTT generation
lib/classes/tts_engines/*.py	Removed per-sentence VTT tracking, added torch.autocast for GPU acceleration, refactored to use common _set_voice/_convert_sml methods
lib/classes/tts_engines/common/utils.py	Replaced _apply_cuda_policy with comprehensive _apply_gpu_policy supporting multiple devices; added _build_vtt_file for batch VTT generation; extracted _set_voice and _convert_sml methods

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.