Mel-Refine: A Plug-and-Play Approach to Refine Mel-Spectrogram in Audio Generation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Guo, Hongming, Fu, Ruibo, Geng, Yizhong, Liu, Shuai, Shi, Shuchen, Wang, Tao, Qiang, Chunyu, Li, Chenxing, Li, Ya, Wen, Zhengqi, Liu, Yukun, Liu, Xuefei
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915059226116096
author Guo, Hongming
Fu, Ruibo
Geng, Yizhong
Liu, Shuai
Shi, Shuchen
Wang, Tao
Qiang, Chunyu
Li, Chenxing
Li, Ya
Wen, Zhengqi
Liu, Yukun
Liu, Xuefei
author_facet Guo, Hongming
Fu, Ruibo
Geng, Yizhong
Liu, Shuai
Shi, Shuchen
Wang, Tao
Qiang, Chunyu
Li, Chenxing
Li, Ya
Wen, Zhengqi
Liu, Yukun
Liu, Xuefei
contents Text-to-audio (TTA) model is capable of generating diverse audio from textual prompts. However, most mainstream TTA models, which predominantly rely on Mel-spectrograms, still face challenges in producing audio with rich content. The intricate details and texture required in Mel-spectrograms for such audio often surpass the models' capacity, leading to outputs that are blurred or lack coherence. In this paper, we begin by investigating the critical role of U-Net in Mel-spectrogram generation. Our analysis shows that in U-Net structure, high-frequency components in skip-connections and the backbone influence texture and detail, while low-frequency components in the backbone are critical for the diffusion denoising process. We further propose ``Mel-Refine'', a plug-and-play approach that enhances Mel-spectrogram texture and detail by adjusting different component weights during inference. Our method requires no additional training or fine-tuning and is fully compatible with any diffusion-based TTA architecture. Experimental results show that our approach boosts performance metrics of the latest TTA model Tango2 by 25\%, demonstrating its effectiveness.
format Preprint
id arxiv_https___arxiv_org_abs_2412_08577
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Mel-Refine: A Plug-and-Play Approach to Refine Mel-Spectrogram in Audio Generation
Guo, Hongming
Fu, Ruibo
Geng, Yizhong
Liu, Shuai
Shi, Shuchen
Wang, Tao
Qiang, Chunyu
Li, Chenxing
Li, Ya
Wen, Zhengqi
Liu, Yukun
Liu, Xuefei
Sound
Multimedia
Audio and Speech Processing
Text-to-audio (TTA) model is capable of generating diverse audio from textual prompts. However, most mainstream TTA models, which predominantly rely on Mel-spectrograms, still face challenges in producing audio with rich content. The intricate details and texture required in Mel-spectrograms for such audio often surpass the models' capacity, leading to outputs that are blurred or lack coherence. In this paper, we begin by investigating the critical role of U-Net in Mel-spectrogram generation. Our analysis shows that in U-Net structure, high-frequency components in skip-connections and the backbone influence texture and detail, while low-frequency components in the backbone are critical for the diffusion denoising process. We further propose ``Mel-Refine'', a plug-and-play approach that enhances Mel-spectrogram texture and detail by adjusting different component weights during inference. Our method requires no additional training or fine-tuning and is fully compatible with any diffusion-based TTA architecture. Experimental results show that our approach boosts performance metrics of the latest TTA model Tango2 by 25\%, demonstrating its effectiveness.
title Mel-Refine: A Plug-and-Play Approach to Refine Mel-Spectrogram in Audio Generation
topic Sound
Multimedia
Audio and Speech Processing
url https://arxiv.org/abs/2412.08577