Our work proposes P-Flow, a fast and data-efficient zero-shot TTS model that uses speech prompts for speaker adaptation. P-Flow comprises a speech-prompted text encoder for speaker adaptation and a flow matching generative decoder for high-quality and fast speech synthesis.

We theoretically establish that our algorithms almost surely converge to locally optimal policies of our safe optimization framework.

Chen et al. [2022] proves that in such a binary-classification problem, an MoE layer converges to an Proof: We will prove this by contradiction. Thus, we show that vanilla MoE does not guarantee convergence with mixture of datasets.

It is desirable to restrict the number of components (i.e., encourage sparsity) for