← All works
Core recurrence

Looped Flow-Matching TTS

Depth through recurrence: Looped transformers for flow-matching TTS

Jiabao Ai, Peng Han, Yuchen Song, Zhengjun Yue

Studies how to organize depth through recurrence in flow-matching text-to-speech by varying the amount, order and placement of weight reuse across seven layouts with the same number of block calls.

Inside the method

18 block calls: e.g. each of 9 blocks × 2 consecutively, or 6 blocks cycled × 3

Simplified conceptual schematic. Consult the paper for the complete architecture.

Recurrence family
Core recurrence
Depth control
Fixed executed depth
KV / state strategy
Not applicable (flow matching)

Reading note

Repeating each of nine blocks twice keeps competitive intelligibility, speaker similarity and quality at 32 steps with 47.1% fewer parameters. Layout rankings change with the sampling budget; middle placement is most consistent.

Sources checked 2026-10-28. This catalog does not imply independent reproduction.

Related works

Cite this work

@misc{loopedtts2026,
  title = {Depth through recurrence: Looped transformers for flow-matching TTS},
  author = {Jiabao Ai and Peng Han and Yuchen Song and Zhengjun Yue},
  year = {2026},
  eprint = {2609.29768},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2609.29768}
}

Suggest a correction on GitHub ↗