Tag
DEFINE is an end-to-end zero-shot TTS framework that decouples speaker identity from accent by conditioning each on separate audio exemplars, with an inference-time guidance weight controlling accent strength on top of F5-TTS using LoRA adaptation.
This paper identifies a confound in best-of-N TTS evaluation where the apparent quality of ASR verifiers depends strongly on which ASR family is used as evaluator. The authors propose cross-family rank ensembles that achieve lower word error rates across multiple evaluators.