Skip to content

Commit 5479e90

Browse files
pre-commit-ci[bot]pre-commit-ci[bot]
authored
[pre-commit.ci] pre-commit autoupdate (#243)
* [pre-commit.ci] pre-commit autoupdate updates: - [github.com/astral-sh/ruff-pre-commit: v0.1.7 → v0.1.8](astral-sh/ruff-pre-commit@v0.1.7...v0.1.8) - [github.com/psf/black: 23.11.0 → 23.12.0](psf/black@23.11.0...23.12.0) * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci --------- Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
1 parent b8fcf2f commit 5479e90

File tree

4 files changed

+67
-39
lines changed

4 files changed

+67
-39
lines changed

‎.pre-commit-config.yaml

+2-2
Original file line numberDiff line numberDiff line change
@@ -7,13 +7,13 @@ repos:
77
- id: trailing-whitespace
88

99
- repo: https://github.com/astral-sh/ruff-pre-commit
10-
rev: v0.1.7
10+
rev: v0.1.8
1111
hooks:
1212
- id: ruff
1313
args: [ --fix ]
1414

1515
- repo: https://github.com/psf/black
16-
rev: 23.11.0
16+
rev: 23.12.0
1717
hooks:
1818
- id: black
1919

‎onnx_infer.py

+2-2
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
import numpy as np
33
Session = OnnxInferenceSession(
44
{
5-
"enc" : "onnx/BertVits2.2PT/BertVits2.2PT_enc_p.onnx",
5+
"enc" : "onnx/BertVits2.2PT/BertVits2.2PT_enc_p.onnx",
66
"emb_g" : "onnx/BertVits2.2PT/BertVits2.2PT_emb.onnx",
77
"dp" : "onnx/BertVits2.2PT/BertVits2.2PT_dp.onnx",
88
"sdp" : "onnx/BertVits2.2PT/BertVits2.2PT_sdp.onnx",
@@ -68,4 +68,4 @@
6868
sid
6969
)
7070

71-
print(audio)
71+
print(audio)

‎onnx_modules/V220_OnnxInference/__init__.py

+63-34
Original file line numberDiff line numberDiff line change
@@ -1,17 +1,20 @@
11
import numpy as np
22
import onnxruntime as ort
33

4+
45
def convert_pad_shape(pad_shape):
56
layer = pad_shape[::-1]
67
pad_shape = [item for sublist in layer for item in sublist]
78
return pad_shape
89

10+
911
def sequence_mask(length, max_length=None):
1012
if max_length is None:
1113
max_length = length.max()
1214
x = np.arange(max_length, dtype=length.dtype)
1315
return np.expand_dims(x, 0) < np.expand_dims(length, 1)
1416

17+
1518
def generate_path(duration, mask):
1619
"""
1720
duration: [b, 1, t_x]
@@ -28,8 +31,9 @@ def generate_path(duration, mask):
2831
path = np.expand_dims(path, 1).transpose(0, 1, 3, 2)
2932
return path
3033

31-
class OnnxInferenceSession():
32-
def __init__(self, path, Providers = ["CPUExecutionProvider"]):
34+
35+
class OnnxInferenceSession:
36+
def __init__(self, path, Providers=["CPUExecutionProvider"]):
3337
self.enc = ort.InferenceSession(path["enc"], providers=Providers)
3438
self.emb_g = ort.InferenceSession(path["emb_g"], providers=Providers)
3539
self.dp = ort.InferenceSession(path["dp"], providers=Providers)
@@ -38,43 +42,56 @@ def __init__(self, path, Providers = ["CPUExecutionProvider"]):
3842
self.dec = ort.InferenceSession(path["dec"], providers=Providers)
3943

4044
def __call__(
41-
self,
42-
seq,
43-
tone,
44-
language,
45-
bert_zh,
46-
bert_jp,
47-
bert_en,
48-
emo,
49-
sid,
50-
seed = 114514,
51-
seq_noise_scale = 0.8,
52-
sdp_noise_scale = 0.6,
53-
length_scale = 1.,
54-
sdp_ratio = 0.
55-
):
56-
g = self.emb_g.run(None, {'sid': sid.astype(np.int64),})[0]
45+
self,
46+
seq,
47+
tone,
48+
language,
49+
bert_zh,
50+
bert_jp,
51+
bert_en,
52+
emo,
53+
sid,
54+
seed=114514,
55+
seq_noise_scale=0.8,
56+
sdp_noise_scale=0.6,
57+
length_scale=1.0,
58+
sdp_ratio=0.0,
59+
):
60+
g = self.emb_g.run(
61+
None,
62+
{
63+
"sid": sid.astype(np.int64),
64+
},
65+
)[0]
5766
g = np.expand_dims(g, -1)
5867
enc_rtn = self.enc.run(
5968
None,
6069
{
61-
"x" : seq.astype(np.int64),
62-
"t" : tone.astype(np.int64),
63-
"language" : language.astype(np.int64),
64-
"bert_0" : bert_zh.astype(np.float32),
65-
"bert_1" : bert_jp.astype(np.float32),
66-
"bert_2" : bert_en.astype(np.float32),
67-
"emo" : emo.astype(np.float32),
68-
"g" : g.astype(np.float32)
69-
})
70+
"x": seq.astype(np.int64),
71+
"t": tone.astype(np.int64),
72+
"language": language.astype(np.int64),
73+
"bert_0": bert_zh.astype(np.float32),
74+
"bert_1": bert_jp.astype(np.float32),
75+
"bert_2": bert_en.astype(np.float32),
76+
"emo": emo.astype(np.float32),
77+
"g": g.astype(np.float32),
78+
},
79+
)
7080
x, m_p, logs_p, x_mask = enc_rtn[0], enc_rtn[1], enc_rtn[2], enc_rtn[3]
7181
np.random.seed(seed)
7282
zinput = np.random.randn(x.shape[0], 2, x.shape[2]) * sdp_noise_scale
73-
logw = self.sdp.run(None, {"x" : x, "x_mask" : x_mask, "zin" : zinput.astype(np.float32), "g" : g})[0] * (sdp_ratio) + \
74-
self.dp.run(None, {"x" : x, "x_mask" : x_mask, "g" : g})[0] * (1 - sdp_ratio)
83+
logw = self.sdp.run(
84+
None, {"x": x, "x_mask": x_mask, "zin": zinput.astype(np.float32), "g": g}
85+
)[0] * (sdp_ratio) + self.dp.run(None, {"x": x, "x_mask": x_mask, "g": g})[
86+
0
87+
] * (
88+
1 - sdp_ratio
89+
)
7590
w = np.exp(logw) * x_mask * length_scale
7691
w_ceil = np.ceil(w)
77-
y_lengths = np.clip(np.sum(w_ceil, (1, 2)), a_min=1., a_max=100000).astype(np.int64)
92+
y_lengths = np.clip(np.sum(w_ceil, (1, 2)), a_min=1.0, a_max=100000).astype(
93+
np.int64
94+
)
7895
y_mask = np.expand_dims(sequence_mask(y_lengths, None), 1)
7996
attn_mask = np.expand_dims(x_mask, 2) * np.expand_dims(y_mask, -1)
8097
attn = generate_path(w_ceil, attn_mask)
@@ -84,9 +101,21 @@ def __call__(
84101
logs_p = np.matmul(attn.squeeze(1), logs_p.transpose(0, 2, 1)).transpose(
85102
0, 2, 1
86103
) # [b, t', t], [b, t, d] -> [b, d, t']
87-
88-
z_p = m_p + np.random.randn(m_p.shape[0], m_p.shape[1], m_p.shape[2]) * np.exp(logs_p) * seq_noise_scale
89104

90-
z = self.flow.run(None, {"z_p" : z_p.astype(np.float32), "y_mask" : y_mask.astype(np.float32), "g": g})[0]
105+
z_p = (
106+
m_p
107+
+ np.random.randn(m_p.shape[0], m_p.shape[1], m_p.shape[2])
108+
* np.exp(logs_p)
109+
* seq_noise_scale
110+
)
111+
112+
z = self.flow.run(
113+
None,
114+
{
115+
"z_p": z_p.astype(np.float32),
116+
"y_mask": y_mask.astype(np.float32),
117+
"g": g,
118+
},
119+
)[0]
91120

92-
return self.dec.run(None, {"z_in" : z.astype(np.float32), "g": g})[0]
121+
return self.dec.run(None, {"z_in": z.astype(np.float32), "g": g})[0]

‎onnx_modules/V220_novq_dev/models_onnx.py

-1
Original file line numberDiff line numberDiff line change
@@ -6,7 +6,6 @@
66
import commons
77
import modules
88
from . import attentions_onnx
9-
from vector_quantize_pytorch import VectorQuantize
109

1110
from torch.nn import Conv1d, ConvTranspose1d, Conv2d
1211
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm

0 commit comments

Comments
 (0)