Added script for testing onnx export.
Browse files- test.ipynb +0 -0
- test.py +84 -0
test.ipynb
CHANGED
The diff for this file is too large to render.
See raw diff
|
|
test.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.environ['TORCH_LOGS'] = '+dynamic'
|
3 |
+
os.environ['TORCH_LOGS'] = '+export'
|
4 |
+
os.environ['TORCHDYNAMO_EXTENDED_DEBUG_GUARD_ADDED']="u0 >= 0"
|
5 |
+
# os.environ['TORCHDYNAMO_EXTENDED_DEBUG_CPP']="1"
|
6 |
+
os.environ['TORCHDYNAMO_EXTENDED_DEBUG_CREATE_SYMBOL']="u0"
|
7 |
+
|
8 |
+
|
9 |
+
from kokoro import phonemize, tokenize, length_to_mask
|
10 |
+
import torch.nn.functional as F
|
11 |
+
from models import build_model
|
12 |
+
import torch
|
13 |
+
device = "cpu" #'cuda' if torch.cuda.is_available() else 'cpu'
|
14 |
+
MODEL = build_model('kokoro-v0_19.pth', device)
|
15 |
+
voicepack = torch.load('voices/af.pt', weights_only=True).to(device)
|
16 |
+
|
17 |
+
model = MODEL
|
18 |
+
speed = 1.
|
19 |
+
|
20 |
+
text = "How could I know? It's an unanswerable question. Like asking an unborn child if they'll lead a good life. They haven't even been born."
|
21 |
+
|
22 |
+
ps = phonemize(text, "a")
|
23 |
+
tokens = tokenize(ps)
|
24 |
+
|
25 |
+
tokens = torch.LongTensor([[0, *tokens, 0]]).to(device)
|
26 |
+
|
27 |
+
class StyleTTS2(torch.nn.Module):
|
28 |
+
def __init__(self, model, voicepack):
|
29 |
+
super().__init__()
|
30 |
+
self.model = model
|
31 |
+
self.voicepack = voicepack
|
32 |
+
|
33 |
+
def forward(self, tokens):
|
34 |
+
speed = 1.
|
35 |
+
# tokens = torch.nn.functional.pad(tokens, (0, 510 - tokens.shape[-1]))
|
36 |
+
device = tokens.device
|
37 |
+
input_lengths = torch.LongTensor([tokens.shape[-1]]).to(device)
|
38 |
+
|
39 |
+
text_mask = length_to_mask(input_lengths).to(device)
|
40 |
+
bert_dur = self.model['bert'](tokens, attention_mask=(~text_mask).int())
|
41 |
+
|
42 |
+
d_en = self.model["bert_encoder"](bert_dur).transpose(-1, -2)
|
43 |
+
|
44 |
+
ref_s = self.voicepack[tokens.shape[1]]
|
45 |
+
s = ref_s[:, 128:]
|
46 |
+
|
47 |
+
d = self.model["predictor"].text_encoder.inference(d_en, s)
|
48 |
+
x, _ = self.model["predictor"].lstm(d)
|
49 |
+
|
50 |
+
duration = self.model["predictor"].duration_proj(x)
|
51 |
+
duration = torch.sigmoid(duration).sum(axis=-1) / speed
|
52 |
+
pred_dur = torch.round(duration).clamp(min=1).long()
|
53 |
+
|
54 |
+
c_start = F.pad(pred_dur,(1,0), "constant").cumsum(dim=1)[0,0:-1]
|
55 |
+
c_end = c_start + pred_dur[0,:]
|
56 |
+
|
57 |
+
torch._check(pred_dur.sum().item()>0, lambda: print(f"Got {pred_dur.sum().item()}"))
|
58 |
+
indices = torch.arange(0, pred_dur.sum().item()).long().to(device)
|
59 |
+
|
60 |
+
pred_aln_trg_list=[]
|
61 |
+
for cs, ce in zip(c_start, c_end):
|
62 |
+
row = torch.where((indices>=cs) & (indices<ce), 1., 0.)
|
63 |
+
pred_aln_trg_list.append(row)
|
64 |
+
pred_aln_trg=torch.vstack(pred_aln_trg_list)
|
65 |
+
|
66 |
+
en = d.transpose(-1, -2) @ pred_aln_trg.unsqueeze(0).to(device)
|
67 |
+
|
68 |
+
F0_pred, N_pred = self.model["predictor"].F0Ntrain(en, s)
|
69 |
+
t_en = self.model["text_encoder"].inference(tokens)
|
70 |
+
asr = t_en @ pred_aln_trg.unsqueeze(0).to(device)
|
71 |
+
return (asr, F0_pred, N_pred, ref_s[:, :128])
|
72 |
+
# output = self.model.decoder(asr, F0_pred, N_pred, ref_s[:, :128]).squeeze().detach().cpu().numpy()
|
73 |
+
|
74 |
+
|
75 |
+
style_model = StyleTTS2(model=model, voicepack=voicepack)
|
76 |
+
(asr, F0_pred, N_pred, ref_s) = style_model(tokens)
|
77 |
+
|
78 |
+
token_len = torch.export.Dim("token_len", min=2, max=510)
|
79 |
+
batch = torch.export.Dim("batch")
|
80 |
+
dynamic_shapes = {"tokens":{0:batch, 1:token_len}}
|
81 |
+
|
82 |
+
# with torch.no_grad():
|
83 |
+
export_mod = torch.export.export(style_model, args=( tokens, ), dynamic_shapes=dynamic_shapes, strict=False)
|
84 |
+
# export_mod = torch.export.export(style_model, args=( tokens, ), strict=False)
|