EzAudio:利用高效扩散变换器增强文本到音频生成功能

Jiarui Hai, Yong Xu, Hao Zhang, Chenxing Li, Helin Wang, Mounya Elhilali, Dong Yu
{"title":"EzAudio:利用高效扩散变换器增强文本到音频生成功能","authors":"Jiarui Hai, Yong Xu, Hao Zhang, Chenxing Li, Helin Wang, Mounya Elhilali, Dong Yu","doi":"arxiv-2409.10819","DOIUrl":null,"url":null,"abstract":"Latent diffusion models have shown promising results in text-to-audio (T2A)\ngeneration tasks, yet previous models have encountered difficulties in\ngeneration quality, computational cost, diffusion sampling, and data\npreparation. In this paper, we introduce EzAudio, a transformer-based T2A\ndiffusion model, to handle these challenges. Our approach includes several key\ninnovations: (1) We build the T2A model on the latent space of a 1D waveform\nVariational Autoencoder (VAE), avoiding the complexities of handling 2D\nspectrogram representations and using an additional neural vocoder. (2) We\ndesign an optimized diffusion transformer architecture specifically tailored\nfor audio latent representations and diffusion modeling, which enhances\nconvergence speed, training stability, and memory usage, making the training\nprocess easier and more efficient. (3) To tackle data scarcity, we adopt a\ndata-efficient training strategy that leverages unlabeled data for learning\nacoustic dependencies, audio caption data annotated by audio-language models\nfor text-to-audio alignment learning, and human-labeled data for fine-tuning.\n(4) We introduce a classifier-free guidance (CFG) rescaling method that\nsimplifies EzAudio by achieving strong prompt alignment while preserving great\naudio quality when using larger CFG scores, eliminating the need to struggle\nwith finding the optimal CFG score to balance this trade-off. EzAudio surpasses\nexisting open-source models in both objective metrics and subjective\nevaluations, delivering realistic listening experiences while maintaining a\nstreamlined model structure, low training costs, and an easy-to-follow training\npipeline. Code, data, and pre-trained models are released at:\nhttps://haidog-yaqub.github.io/EzAudio-Page/.","PeriodicalId":501284,"journal":{"name":"arXiv - EE - Audio and Speech Processing","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2024-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"EzAudio: Enhancing Text-to-Audio Generation with Efficient Diffusion Transformer\",\"authors\":\"Jiarui Hai, Yong Xu, Hao Zhang, Chenxing Li, Helin Wang, Mounya Elhilali, Dong Yu\",\"doi\":\"arxiv-2409.10819\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Latent diffusion models have shown promising results in text-to-audio (T2A)\\ngeneration tasks, yet previous models have encountered difficulties in\\ngeneration quality, computational cost, diffusion sampling, and data\\npreparation. In this paper, we introduce EzAudio, a transformer-based T2A\\ndiffusion model, to handle these challenges. Our approach includes several key\\ninnovations: (1) We build the T2A model on the latent space of a 1D waveform\\nVariational Autoencoder (VAE), avoiding the complexities of handling 2D\\nspectrogram representations and using an additional neural vocoder. (2) We\\ndesign an optimized diffusion transformer architecture specifically tailored\\nfor audio latent representations and diffusion modeling, which enhances\\nconvergence speed, training stability, and memory usage, making the training\\nprocess easier and more efficient. (3) To tackle data scarcity, we adopt a\\ndata-efficient training strategy that leverages unlabeled data for learning\\nacoustic dependencies, audio caption data annotated by audio-language models\\nfor text-to-audio alignment learning, and human-labeled data for fine-tuning.\\n(4) We introduce a classifier-free guidance (CFG) rescaling method that\\nsimplifies EzAudio by achieving strong prompt alignment while preserving great\\naudio quality when using larger CFG scores, eliminating the need to struggle\\nwith finding the optimal CFG score to balance this trade-off. EzAudio surpasses\\nexisting open-source models in both objective metrics and subjective\\nevaluations, delivering realistic listening experiences while maintaining a\\nstreamlined model structure, low training costs, and an easy-to-follow training\\npipeline. Code, data, and pre-trained models are released at:\\nhttps://haidog-yaqub.github.io/EzAudio-Page/.\",\"PeriodicalId\":501284,\"journal\":{\"name\":\"arXiv - EE - Audio and Speech Processing\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-09-17\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"arXiv - EE - Audio and Speech Processing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/arxiv-2409.10819\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"arXiv - EE - Audio and Speech Processing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/arxiv-2409.10819","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

潜在扩散模型在文本到音频(T2A)生成任务中取得了可喜的成果,但以前的模型在生成质量、计算成本、扩散采样和数据准备方面遇到了困难。在本文中,我们介绍了基于变压器的 T2A 扩散模型 EzAudio,以应对这些挑战。我们的方法包括几项关键创新:(1)我们在一维波形变异自动编码器(VAE)的潜空间上建立 T2A 模型,避免了处理二维频谱图表示的复杂性,并使用了额外的神经声码器。(2) 专门针对音频潜在表示和扩散建模设计了优化的扩散变换器架构,提高了收敛速度、训练稳定性和内存使用率,使训练过程更简单、更高效。(3) 为了解决数据稀缺的问题,我们采用了一种数据高效的训练策略,即利用未标记数据学习声学依赖关系,利用音频语言模型注释的音频标题数据学习文本到音频的配准,以及利用人类标记数据进行微调。(4) 我们引入了一种无分类器引导(CFG)重缩放方法,该方法简化了 EzAudio,在使用较大的 CFG 分数时,既能实现较强的提示对齐,又能保持较高的音频质量,从而无需费力寻找最佳 CFG 分数来平衡这种权衡。EzAudio 在客观指标和主观评价方面都超越了现有的开源模型,在提供逼真的听觉体验的同时,还保持了精简的模型结构、较低的训练成本和简单易学的训练管道。代码、数据和预训练模型的发布网址为:https://haidog-yaqub.github.io/EzAudio-Page/。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
EzAudio: Enhancing Text-to-Audio Generation with Efficient Diffusion Transformer
Latent diffusion models have shown promising results in text-to-audio (T2A) generation tasks, yet previous models have encountered difficulties in generation quality, computational cost, diffusion sampling, and data preparation. In this paper, we introduce EzAudio, a transformer-based T2A diffusion model, to handle these challenges. Our approach includes several key innovations: (1) We build the T2A model on the latent space of a 1D waveform Variational Autoencoder (VAE), avoiding the complexities of handling 2D spectrogram representations and using an additional neural vocoder. (2) We design an optimized diffusion transformer architecture specifically tailored for audio latent representations and diffusion modeling, which enhances convergence speed, training stability, and memory usage, making the training process easier and more efficient. (3) To tackle data scarcity, we adopt a data-efficient training strategy that leverages unlabeled data for learning acoustic dependencies, audio caption data annotated by audio-language models for text-to-audio alignment learning, and human-labeled data for fine-tuning. (4) We introduce a classifier-free guidance (CFG) rescaling method that simplifies EzAudio by achieving strong prompt alignment while preserving great audio quality when using larger CFG scores, eliminating the need to struggle with finding the optimal CFG score to balance this trade-off. EzAudio surpasses existing open-source models in both objective metrics and subjective evaluations, delivering realistic listening experiences while maintaining a streamlined model structure, low training costs, and an easy-to-follow training pipeline. Code, data, and pre-trained models are released at: https://haidog-yaqub.github.io/EzAudio-Page/.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信