Scholarlink

The exponential growth of audio data has necessitated innovative solutions for the real-time upload and storage of large audio volumes within the constraints of limited network bandwidth and storage capacity. This paper presents a high-fidelity neural audio compression technique designed to compress 44.1 kHz audio at an enhanced compression ratio. This is achieved through the integration of advanced vector quantization methods with Generative Adversarial Networks and reconstructive loss functions. Experimental results indicate that the proposed algorithm outperforms both deep learning-based and traditional audio compression methods, exhibiting superior reconstruction quality, higher compression rates, and enhanced audio fidelity.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Designing an Audio CODEC with Synthesizing Generative Adversarial Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Asish Debnath,
  • Uttam Kr. Mondal

摘要

The exponential growth of audio data has necessitated innovative solutions for the real-time upload and storage of large audio volumes within the constraints of limited network bandwidth and storage capacity. This paper presents a high-fidelity neural audio compression technique designed to compress 44.1 kHz audio at an enhanced compression ratio. This is achieved through the integration of advanced vector quantization methods with Generative Adversarial Networks and reconstructive loss functions. Experimental results indicate that the proposed algorithm outperforms both deep learning-based and traditional audio compression methods, exhibiting superior reconstruction quality, higher compression rates, and enhanced audio fidelity.