Release new vocoder: firefly-gan-base
- EVA-GAN architecture (convnext encoder + hifigan), but without loss balancer
- Trained on 24xA800 GPUs for 1 week
- HiFi-16kh dataset
Check scripts/test_firefly_gan.sh to know how to use this model.
This model is released under CC-BY-NC-SA 4.0 license, please read it before you download.