Model
HiFi-GAN v1 PyTorch checkpoint trained on 8GPU with AMP on LJSpeech-1.1 (22kHz), fine-tuned on FastPitch outputs.
Use the NGC CLI to download:
Copied!
1 Version
21.08.0_ampSelected06/30/2022 11:23 PM UTC53.23 MBAccuracy: 010000 EpochsBatch Size: 0GPU: V100 Copied!
21.08.0_ampSelected
06/30/2022 11:23 PM UTC53.23 MBAccuracy: 010000 EpochsBatch Size: 0GPU: V100
Copied!
architecture
| Key | Value |
|---|---|
| task | training |
| type | v1 22kHz |
performance
| Key | Value |
|---|---|
| mel_loss | 25.251 |
training
| Key | Value |
|---|---|
| sampling_rate | 22050 |
| bs | 128 |
| lr_decay | 0.9998 |
| lr | 0.0003 |
| training_precision | AMP |
| epochs | 10000 |
| dataset | LJSpeech-1.1 |
| fine_tune_lr_factor | 3 |