v2.4

• Update checkpoints with 5M training steps. This is the final release of BigVGAN-v2 pretrained checkpoints.
• Robust MelDataset with heterogeneous waveform file format support and on-the-fly resampling during training: PR#8
• Bug fixes: PR#7 PR#9

v2.3

• General refactor and code improvements for improved readability: PR#4 PR#5
• Fully fused CUDA kernel of anti-alised activation (upsampling + activation + downsampling) with inference speed benchmark: PR#6

v2.2

• Integrate local gradio demo: PR#2

v2.1

• Hugging Face Hub integration: PR#1
• Refactor models.py into bigvgan.py, discriminators.py, and loss.py: PR#1

v2

Commit

• Custom CUDA kernel for inference
• Improved discriminator and loss
• New checkpoints trained on more diverse data

v1

Original v1 release of BigVGAN