https://hgpu.org/?p=27588
Efficient Incremental Text-to-Speech on GPUs