文本转语音中用于韵律和声学建模的归一化流与扩散模型比较
Comparing normalizing flows and diffusion models for prosody and acoustic modelling in text-to-speech
AI总结：
本文比较了文本转语音中基于 L1/L2 损失的传统方法与 Normalizing Flows 和 Diffusion 模型在韵律和 mel-spectrogram 预测上的表现，发现流模型在频谱图预测上最优，且流与扩散韵律预测器均显著优于 L2 模型。
摘要：
神经文本转语音系统通常基于 L1/L2 损失进行优化，这会对目标数据空间的分布做出强假设。为改善这些假设，最近提出了 Normalizing Flows 和 Diffusion Probabilistic Models 作为替代方案。本文针对文本转语音合成中的韵律和 mel-spectrogram 预测任务，将传统的基于 L1/L2 的方法与扩散和流方法进行了比较。我们使用韵律模型生成 log-f0 和时长特征，以此作为生成 mel-spectrograms 的声学模型的条件。实验结果表明，基于流的模型在频谱图预测上取得最佳性能，优于等效的扩散和 L1 模型。同时，基于扩散和流的韵律预测器均比典型的 L2 训练韵律模型有显著提升。
英文摘要：
Neural text-to-speech systems are often optimized on L1/L2 losses, which make strong assumptions about the distributions of the target data space. Aiming to improve those assumptions, Normalizing Flows and Diffusion Probabilistic Models were recently proposed as alternatives. In this paper, we compare traditional L1/L2-based approaches to diffusion and flow-based approaches for the tasks of prosody and mel-spectrogram prediction for text-to-speech synthesis. We use a prosody model to generate log-f0 and duration features, which are used to condition an acoustic model that generates mel-spectrograms. Experimental results demonstrate that the flow-based model achieves the best performance for spectrogram prediction, improving over equivalent diffusion and L1 models. Meanwhile, both diffusion and flow-based prosody predictors result in significant improvements over a typical L2-trained prosody models.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
