SEAM: Spectral Embedding Alignment and Merging for Low-Resource TTS