storytold/vits-finetuning
2
0
Fork 0
mirror of https://github.com/storytold/vits-finetuning.git synced 2026-10-09 00:09:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make easier to train on max sample length = 10s

Browse Source
This commit is contained in:
ZDisket
2023-05-26 16:49:23 -03:00
parent 2a263444a1
commit ca694d0190
4 changed files with 7 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
configs/ljs_li44_tmbert_ft_small_s1_arpa.json
+2 -2
View File
@@ -8,8 +8,8 @@
"learning_rate": 2e-4, "learning_rate": 2e-4,
"betas": [0.8, 0.99], "betas": [0.8, 0.99],
"eps": 1e-9, "eps": 1e-9,
"batch_size": 16, "batch_size": 12,
"grad_acc_steps": 2, "grad_acc_steps": 3,
"fp16_run": true, "fp16_run": true,
"lr_decay": 0.999875, "lr_decay": 0.999875,
"segment_size": 8192, "segment_size": 8192,
data_utils.py
+1 -1
View File
@@ -58,7 +58,7 @@ class TextAudioLoader(torch.utils.data.Dataset):
continue continue
au_len = librosa.get_duration(filename=audiopath) au_len = librosa.get_duration(filename=audiopath)
if au_len > 13.0: if au_len > 10.0:
n_fau += 1 n_fau += 1
continue continue
preprocess.py
+1 -1
View File
@@ -24,7 +24,7 @@ if __name__ == '__main__':
test_sentences = ["The quick brown fox jumps over the lazy dog", test_sentences = ["The quick brown fox jumps over the lazy dog",
"In a galaxy far, far away, a young hero embarks on an epic adventure", "In a galaxy far, far away, a young hero embarks on an epic adventure",
"Ladies and gentlemen, welcome to the annual science fair!", "Ladies and gentlemen, welcome to the annual science fair!",
"The chef skillfully prepares a delicious gourmet meal with fresh ingredients and exquisite flavors", "Peter Piper picked a peck of pickled peppers. How many pickled peppers did Peter Piper pick?",
"The crowd erupted in cheers as the team scored the winning goal in the final seconds of the match"] "The crowd erupted in cheers as the team scored the winning goal in the final seconds of the match"]
if len(args.test) > 1: if len(args.test) > 1:
train.py
+3 -1
View File
@@ -394,7 +394,9 @@ def evaluate(hps, generator, eval_loader, writer_eval):
t_bert_lens = torch.LongTensor([t_bert.size(1)]).cuda(0) t_bert_lens = torch.LongTensor([t_bert.size(1)]).cuda(0)
t_bert = t_bert.cuda(0) t_bert = t_bert.cuda(0)
audio, attn, t_mask, _ = generator.module.infer(t_text_norm, t_text_lengths, t_moji, t_bert, t_bert_lens, noise_scale=.667, noise_scale_w=0.8, length_scale=1.0) with torch.no_grad():
audio, attn, t_mask, _ = generator.module.infer(t_text_norm, t_text_lengths, t_moji, t_bert, t_bert_lens, noise_scale=.667, noise_scale_w=0.8, length_scale=1.0)
test_audio_lengths = t_mask.sum([1,2]).long() * hps.data.hop_length test_audio_lengths = t_mask.sum([1,2]).long() * hps.data.hop_length
y_test_mel = mel_spectrogram_torch( y_test_mel = mel_spectrogram_torch(
audio.squeeze(1).float(), audio.squeeze(1).float(),