• Use piper1-gpl library instead of piper binary
• Use sentence-stream library instead of internal code
• Add --use-cuda to enable GPU acceleration (requires onnxruntime-gpu)
• Ignore file sizes and hashes when downloading voices
• Default streaming to be on (remove --streaming) and add --no-streaming to disable
• Remove --piper and --max-piper-procs (no longer needed)
• Add Docker build here
• Add publish workflow
• Add alias --noise-w-scale for --noise-w to align with piper1-gpl