BemaGANv2: A Tutorial and Comparative Survey of GAN-based Vocoders for Long-Term Audio Generation

2025-06-11

AI Summary

To address inaccurate periodicity modeling and insufficient long-range dependency capture in GAN-based vocoders for long-duration speech synthesis, this paper proposes BemaGANv2—a novel GAN vocoder designed for high-fidelity, long-audio waveform generation. Its core innovations include: (1) the Anti-aliased Multi-Periodicity (AMP) generator module—incorporating Snake activation—to enforce consistency between fundamental frequency and harmonic periodicities; and (2) a Multi-Envelope Discriminator (MED) jointly deployed with the Multi-Receptive-Field Discriminator (MRD) to systematically enhance long-term structural modeling. Comprehensive multi-scale evaluation (FAD, SSIM, PLCC, MCD, MOS, SMOS) demonstrates that BemaGANv2 significantly outperforms state-of-the-art GAN vocoders in both objective metrics and subjective MOS scores. The code and pre-trained models are fully open-sourced.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

This paper presents a tutorial-style survey and implementation guide of BemaGANv2, an advanced GAN-based vocoder designed for high-fidelity and long-term audio generation. Built upon the original BemaGAN architecture, BemaGANv2 incorporates major architectural innovations by replacing traditional ResBlocks in the generator with the Anti-aliased Multi-Periodicity composition (AMP) module, which internally applies the Snake activation function to better model periodic structures. In the discriminator framework, we integrate the Multi-Envelope Discriminator (MED), a novel architecture we originally proposed, to extract rich temporal envelope features crucial for periodicity detection. Coupled with the Multi-Resolution Discriminator (MRD), this combination enables more accurate modeling of long-range dependencies in audio. We systematically evaluate various discriminator configurations, including MSD + MED, MSD + MRD, and MPD + MED + MRD, using objective metrics (FAD, SSIM, PLCC, MCD) and subjective evaluations (MOS, SMOS). This paper also provides a comprehensive tutorial on the model architecture, training methodology, and implementation to promote reproducibility. The code and pre-trained models are available at: https://github.com/dinhoitt/BemaGANv2.

Problem

• Improving GAN-based vocoders for high-fidelity audio generation
• Enhancing long-term audio modeling with novel AMP modules
• Evaluating discriminator configurations for better periodicity detection

Innovation

• AMP module replaces ResBlocks in generator
• MED discriminator extracts temporal envelope features
• Combines MED and MRD for long-range dependencies