研究人员开发了一系列生成语音增强模型，以PASE为起点，该模型利用WavLM的语音学先验来减少幻觉。后续的StuPASE和UniPASE在此基础上进行构建。StuPASE通过用流匹配方法替换其生成模块来增强感知质量并处理严重噪声，而UniPASE则使用称为DeWavLM-Omni的统一表示模块将该框架扩展到跨多个采样率的通用语音增强。与以前的方法相比，这些模型旨在以显着降低的语言和声学幻觉实现录音室级的语音恢复。 AI
影响 这些模型推动了音频处理领域生成式AI的能力，有可能改进语音助手和音频恢复工具等应用。
排序理由 多篇研究论文详细介绍了生成语音增强模型的进展。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2511.13300v1 Announce Type: cross Abstract: Generative models have shown remarkable performance in speech enhancement (SE), achieving superior perceptual quality over traditional discriminative approaches. However, existing generative SE approaches often overlook the risk o…
arXiv:2603.09234v2 Announce Type: cross Abstract: Achieving high perceptual quality without hallucination remains a challenge in generative speech enhancement (SE). A representative approach, PASE, is robust to hallucination but has limited perceptual quality under adverse condit…
arXiv:2604.14606v2 Announce Type: cross Abstract: Universal speech enhancement (USE) aims to restore speech signals from diverse distortions across multiple sampling rates. We propose UniPASE, an extension of the low-hallucination PASE framework tailored for USE. At its core is D…