TTML Lyrics Studio v2.1.0
Highlights & Fixes
- Robust Audio Alignment (Issue #1):
- Fixed WhisperX alignment audio cutoffs on tracks longer than 5 minutes by using explicit batch chunking and tuned Voice Activity Detection (VAD).
- Preserved verbatim user lyrics words in word-level and syllable-level alignments, preventing Whisper auto-corrections and hallucinations on regional languages, dialects, and Creole.
- Added graceful Wav2Vec2 CTC fallback for languages without pre-trained phoneme alignment models.
- Python & CUDA Hardware Compatibility:
- install.bat now verifies that Python is version 3.10 to 3.13 and blocks unsupported Python 3.14+.
- Updated PyTorch to CUDA 12.6, providing out-of-the-box hardware acceleration for NVIDIA RTX 20xx, 30xx, 40xx, and 50xx series GPUs.
- Automatically installs WhisperX in the isolated virtual environment (.venv).
- Dashboard & Autosave Recovery:
- Recent Projects dashboard with one-click project reopening, recovery from project-bound autosaves, and direct .lrcproj drag-and-drop.
- Multi-track status overview and album lane previews.
- Studio Documentation:
- Complete overhaul of the README with high-resolution 1080p workspace screenshots and keyboard shortcut reference guide.
Quick Start (Windows)
- Download and extract LRCGen-v2.1.0-Windows.zip.
- Double-click install.bat to install dependencies (Python 3.10 to 3.13 required).
- Double-click start.bat to launch the studio in your browser.