pinned Running on Zero TADA — Text-Acoustic Dual Alignment for Speech 🗯 Speech generation from text and acoustic reference