name: Multi-Speaker Dialogue Generation
description: The Multi-Speaker Dialogue Generation task evaluates the model’s ability to generate speech in multi-turn dialogue scenarios. We report results under two input settings: with and without instructions. This comparison demonstrates the model’s natural controllability through contextual scene understanding, as well as its guided controllability when provided with instructions. It should be noted that the compared baseline models do not provide explicit controllability, and are therefore evaluated only under their default generation settings.
models:
  - MOSS-TTSD-V1(no instruction)
  - SoulX-Podcast(no instruction)
  - Ours(no instruction)
  - Ours(with instruction)