arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2309.08051cs.SDcs.AIcs.MMeess.AS
收藏

检索增强的文本到音频生成

Retrieval-Augmented Text-to-Audio Generation

University of Surrey(萨里大学)

Yi Yuan, Haohe Liu, Xubo Liu, Qiushi Huang, Mark D. Plumbley, Wenwu Wang

AI总结:

针对文本到音频生成中因数据类别分布不平衡导致长尾类别生成性能差的问题,提出检索增强方法 Re-AudioLDM,利用 CLAP 模型检索相关文本-音频对作为附加条件指导生成,在 AudioCaps 上取得 1.37 的最优 FAD。

摘要:

尽管文本到音频(TTA)生成近期取得进展,我们展示了如 AudioLDM 等最先进模型,在 AudioCaps 等类别分布不平衡的数据集上训练时,其生成性能存在偏差。具体而言,它们擅长生成常见音频类别,但在罕见类别上表现不佳,从而降低了整体生成性能。我们将此问题称为长尾文本到音频生成。为解决此问题,我们提出一种简单的 TTA 模型检索增强方法。具体而言,给定输入文本提示,我们首先利用对比语言音频预训练(CLAP)模型检索相关的文本-音频对。检索到的音频-文本数据特征随后被用作附加条件,以指导 TTA 模型的学习。我们用提出的方法增强 AudioLDM,并将所得增强系统记为 Re-AudioLDM。在 AudioCaps 数据集上,Re-AudioLDM 实现了 1.37 的最先进 Frechet Audio Distance (FAD),大幅超越现有方法。此外,我们表明 Re-AudioLDM 能为复杂场景、罕见音频类别甚至未见音频类型生成逼真音频,表明其在 TTA 任务中的潜力。

英文摘要:

Despite recent progress in text-to-audio (TTA) generation, we show that the state-of-the-art models, such as AudioLDM, trained on datasets with an imbalanced class distribution, such as AudioCaps, are biased in their generation performance. Specifically, they excel in generating common audio classes while underperforming in the rare ones, thus degrading the overall generation performance. We refer to this problem as long-tailed text-to-audio generation. To address this issue, we propose a simple retrieval-augmented approach for TTA models. Specifically, given an input text prompt, we first leverage a Contrastive Language Audio Pretraining (CLAP) model to retrieve relevant text-audio pairs. The features of the retrieved audio-text data are then used as additional conditions to guide the learning of TTA models. We enhance AudioLDM with our proposed approach and denote the resulting augmented system as Re-AudioLDM. On the AudioCaps dataset, Re-AudioLDM achieves a state-of-the-art Frechet Audio Distance (FAD) of 1.37, outperforming the existing approaches by a large margin. Furthermore, we show that Re-AudioLDM can generate realistic audio for complex scenes, rare audio classes, and even unseen audio types, indicating its potential in TTA tasks.

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。

↑