RAF: Relativistic Adversarial Feedback For Universal Speech Synthesis
RAF:用于通用语音合成的相对论对抗反馈
机构 * Korea Advanced Institute of Science and Technology (KAIST)(韩国科学技术院)
AI总结 提出相对论对抗反馈(RAF)训练目标,通过自监督语音模型和相对论配对改进GAN声码器的域内保真度和泛化能力,在参数减少88%的情况下超越LSGAN训练的BigVGAN。
Comments Accepted to Interspeech 2026 Long paper track. Code: https://github.com/infected4098/Relativistic-Adversarial-Feedback