Auto-transcribe selected voice with CPU-only Whisper to pre-fill the Higgs reference transcript

Add Higgs Audio v3 (4B) as a selectable voice backend alongside DramaBox

Initial DramaBox TTS space