pinned Running TADA โ Text-Acoustic Dual Alignment for Speech ๐ฏ Speech generation from text and acoustic reference