Saved in:
Bibliographic Details
Main Authors: Qiu, Kai, Li, Xiang, Chen, Hao, Sun, Jie, Wang, Jinglu, Lin, Zhe, Savvides, Marios, Raj, Bhiksha
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2408.09027
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915066517913600
author Qiu, Kai
Li, Xiang
Chen, Hao
Sun, Jie
Wang, Jinglu
Lin, Zhe
Savvides, Marios
Raj, Bhiksha
author_facet Qiu, Kai
Li, Xiang
Chen, Hao
Sun, Jie
Wang, Jinglu
Lin, Zhe
Savvides, Marios
Raj, Bhiksha
contents Audio generation has achieved remarkable progress with the advance of sophisticated generative models, such as diffusion models (DMs) and autoregressive (AR) models. However, due to the naturally significant sequence length of audio, the efficiency of audio generation remains an essential issue to be addressed, especially for AR models that are incorporated in large language models (LLMs). In this paper, we analyze the token length of audio tokenization and propose a novel \textbf{S}cale-level \textbf{A}udio \textbf{T}okenizer (SAT), with improved residual quantization. Based on SAT, a scale-level \textbf{A}coustic \textbf{A}uto\textbf{R}egressive (AAR) modeling framework is further proposed, which shifts the next-token AR prediction to next-scale AR prediction, significantly reducing the training cost and inference time. To validate the effectiveness of the proposed approach, we comprehensively analyze design choices and demonstrate the proposed AAR framework achieves a remarkable \textbf{35}$\times$ faster inference speed and +\textbf{1.33} Fréchet Audio Distance (FAD) against baselines on the AudioSet benchmark. Code: \url{https://github.com/qiuk2/AAR}.
format Preprint
id arxiv_https___arxiv_org_abs_2408_09027
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Efficient Autoregressive Audio Modeling via Next-Scale Prediction
Qiu, Kai
Li, Xiang
Chen, Hao
Sun, Jie
Wang, Jinglu
Lin, Zhe
Savvides, Marios
Raj, Bhiksha
Sound
Artificial Intelligence
Audio and Speech Processing
Audio generation has achieved remarkable progress with the advance of sophisticated generative models, such as diffusion models (DMs) and autoregressive (AR) models. However, due to the naturally significant sequence length of audio, the efficiency of audio generation remains an essential issue to be addressed, especially for AR models that are incorporated in large language models (LLMs). In this paper, we analyze the token length of audio tokenization and propose a novel \textbf{S}cale-level \textbf{A}udio \textbf{T}okenizer (SAT), with improved residual quantization. Based on SAT, a scale-level \textbf{A}coustic \textbf{A}uto\textbf{R}egressive (AAR) modeling framework is further proposed, which shifts the next-token AR prediction to next-scale AR prediction, significantly reducing the training cost and inference time. To validate the effectiveness of the proposed approach, we comprehensively analyze design choices and demonstrate the proposed AAR framework achieves a remarkable \textbf{35}$\times$ faster inference speed and +\textbf{1.33} Fréchet Audio Distance (FAD) against baselines on the AudioSet benchmark. Code: \url{https://github.com/qiuk2/AAR}.
title Efficient Autoregressive Audio Modeling via Next-Scale Prediction
topic Sound
Artificial Intelligence
Audio and Speech Processing
url https://arxiv.org/abs/2408.09027