mrq/DL-Art-School
1
3
Fork 5
Code Issues Pull Requests Packages Projects Releases Wiki Activity

another fix

Browse Source
This commit is contained in:
James Betker 2022-04-11 12:29:43 -06:00
parent f2c172291f
commit 6dea7da7a8
1 changed files with 1 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
codes/trainer/eval/audio_diffusion_fid.py
View File

@ -18,8 +18,7 @@ from models.audio.tts.tacotron2 import text_to_sequence
from scripts.audio.gen.speech_synthesis_utils import load_discrete_vocoder_diffuser, wav_to_mel, load_speech_dvae, \
convert_mel_to_codes, load_univnet_vocoder, wav_to_univnet_mel
from trainer.injectors.audio_injectors import denormalize_tacotron_mel
from utils.util import ceil_multiple, opt_get
from utils.util import ceil_multiple, opt_get, load_model_from_config
class AudioDiffusionFid(evaluator.Evaluator):
"""
@ -280,7 +279,6 @@ if __name__ == '__main__':
if __name__ == '__main__':
from utils.util import load_model_from_config
# 34k; no conditioning_free: {'frechet_distance': tensor(1.4559, device='cuda:0', dtype=torch.float64), 'intelligibility_loss': tensor(151.9112, device='cuda:0')}
# 34k; conditioning_free: {'frechet_distance': tensor(1.4059, device='cuda:0', dtype=torch.float64), 'intelligibility_loss': tensor(118.3377, device='cuda:0')}
diffusion = load_model_from_config('X:\\dlas\\experiments\\train_diffusion_tts_mel_flat_autoregressive_inputs.yml', 'generator',