BemaGANv2: A Vocoder with Superior Periodicity Capture for Long-Term Audio Generation

Taesoo Park, Mungwi Jeong, Mingyu Park, Narae Kim, Junyoung Kim, Soonchul Kwon, Jisang Yoo

In our paper, We proposed BemaGANv2: Discriminator Combination Strategies for GAN-based Vocoders in Long-Term Audio Generation
We provide our implementation and pretrained models as open source in this repository.

We provide our sample testset. demoweb

Abstract :

This paper presents BemaGANv2, an advanced GAN-based vocoder designed for high-fidelity and long-term audio generation, with a focus on systematic evaluation of discriminator combination strategies. Long-term audio generation is critical for applications in Text-to-Music (TTM) and Text-to-Audio (TTA)systems, where maintaining temporal coherence, prosodic consistency, and harmonic structure over extended durations remains a significant challenge. 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 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 Multi-Scale Discriminator (MSD) + MED, MSD + MRD, and Multi-Period Discriminator (MPD) + MED + MRD, using objective metrics (Fréchet Audio Distance (FAD), Structural Similarity Index (SSIM), Pearson Correlation Coefficient (PCC), Mel-Cepstral Distortion (MCD), MultiResolution STFT (M-STFT), Periodicity error (Periodicity)) and subjective evaluations (MOS, SMOS). To support reproducibility, we provide detailed architectural descriptions, training configurations, and complete implementation details. The code, pre-trained models, and audio demo samples are available at: https://github.com/dinhoitt/BemaGANv2.

---

Our settings

1. python = 3.9
2. Clone this repository.
3. Install requirements.txt

pip install -r requirements.txt

dataset

Download and extract the LJ Speech dataset.

And move all wav files to LJSpeech-1.1/wavs0

Run the code below.

import os
import glob

input_dir = "/Your path/LJSpeech-1.1/wavs0"
output_dir = "/Your path/LJSpeech-1.1/wavs"
os.makedirs(output_dir, exist_ok=True)

for file in glob.glob(f"{input_dir}/*.wav"):
    filename = os.path.basename(file)
    out_path = os.path.join(output_dir, filename)
    !ffmpeg -y -loglevel panic -i "{file}" -ar 24000 "{out_path}"

---

Training

python train.py --config config_v1.json

Checkpoints and copy of the configuration file are saved in cp_BemaGanv2_MED_MRD directory by default.
You can change the path by adding --checkpoint_path option.

Training gen loss total & mel spec error

validation mel spec error

Inference from wav file

1. Make test_files directory and copy wav files into the directory.
2. Run the following command.

   python inference.py --checkpoint_file [generator checkpoint file path]
   

Generated wav files are saved in generated_files by default.
You can change the path by adding --input_wavs_dir , --output_dir option.

---

---

This repository was created with the participation of Capstone Design members from Kwangwoon University.

Contributors:

• Taesoo Park
• Narae Kim
• Mungwi Jeong
• Mingyu Park
• Junyoung Kim