[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](https://github.com/astral-sh/ruff-pre-commit/compare/v0.1.7...v0.1.8) - [github.com/psf/black: 23.11.0 → 23.12.0](https://github.com/psf/black/compare/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>
This commit is contained in:
pre-commit-ci[bot]
committed by
GitHub
GitHub
parent
b8fcf2f66e
commit
5479e9039d
@@ -7,13 +7,13 @@ repos:
|
||||
- id: trailing-whitespace
|
||||
|
||||
- repo: https://github.com/astral-sh/ruff-pre-commit
|
||||
rev: v0.1.7
|
||||
rev: v0.1.8
|
||||
hooks:
|
||||
- id: ruff
|
||||
args: [ --fix ]
|
||||
|
||||
- repo: https://github.com/psf/black
|
||||
rev: 23.11.0
|
||||
rev: 23.12.0
|
||||
hooks:
|
||||
- id: black
|
||||
|
||||
|
||||
@@ -1,17 +1,20 @@
|
||||
import numpy as np
|
||||
import onnxruntime as ort
|
||||
|
||||
|
||||
def convert_pad_shape(pad_shape):
|
||||
layer = pad_shape[::-1]
|
||||
pad_shape = [item for sublist in layer for item in sublist]
|
||||
return pad_shape
|
||||
|
||||
|
||||
def sequence_mask(length, max_length=None):
|
||||
if max_length is None:
|
||||
max_length = length.max()
|
||||
x = np.arange(max_length, dtype=length.dtype)
|
||||
return np.expand_dims(x, 0) < np.expand_dims(length, 1)
|
||||
|
||||
|
||||
def generate_path(duration, mask):
|
||||
"""
|
||||
duration: [b, 1, t_x]
|
||||
@@ -28,8 +31,9 @@ def generate_path(duration, mask):
|
||||
path = np.expand_dims(path, 1).transpose(0, 1, 3, 2)
|
||||
return path
|
||||
|
||||
class OnnxInferenceSession():
|
||||
def __init__(self, path, Providers = ["CPUExecutionProvider"]):
|
||||
|
||||
class OnnxInferenceSession:
|
||||
def __init__(self, path, Providers=["CPUExecutionProvider"]):
|
||||
self.enc = ort.InferenceSession(path["enc"], providers=Providers)
|
||||
self.emb_g = ort.InferenceSession(path["emb_g"], providers=Providers)
|
||||
self.dp = ort.InferenceSession(path["dp"], providers=Providers)
|
||||
@@ -47,34 +51,47 @@ class OnnxInferenceSession():
|
||||
bert_en,
|
||||
emo,
|
||||
sid,
|
||||
seed = 114514,
|
||||
seq_noise_scale = 0.8,
|
||||
sdp_noise_scale = 0.6,
|
||||
length_scale = 1.,
|
||||
sdp_ratio = 0.
|
||||
seed=114514,
|
||||
seq_noise_scale=0.8,
|
||||
sdp_noise_scale=0.6,
|
||||
length_scale=1.0,
|
||||
sdp_ratio=0.0,
|
||||
):
|
||||
g = self.emb_g.run(None, {'sid': sid.astype(np.int64),})[0]
|
||||
g = self.emb_g.run(
|
||||
None,
|
||||
{
|
||||
"sid": sid.astype(np.int64),
|
||||
},
|
||||
)[0]
|
||||
g = np.expand_dims(g, -1)
|
||||
enc_rtn = self.enc.run(
|
||||
None,
|
||||
{
|
||||
"x" : seq.astype(np.int64),
|
||||
"t" : tone.astype(np.int64),
|
||||
"language" : language.astype(np.int64),
|
||||
"bert_0" : bert_zh.astype(np.float32),
|
||||
"bert_1" : bert_jp.astype(np.float32),
|
||||
"bert_2" : bert_en.astype(np.float32),
|
||||
"emo" : emo.astype(np.float32),
|
||||
"g" : g.astype(np.float32)
|
||||
})
|
||||
"x": seq.astype(np.int64),
|
||||
"t": tone.astype(np.int64),
|
||||
"language": language.astype(np.int64),
|
||||
"bert_0": bert_zh.astype(np.float32),
|
||||
"bert_1": bert_jp.astype(np.float32),
|
||||
"bert_2": bert_en.astype(np.float32),
|
||||
"emo": emo.astype(np.float32),
|
||||
"g": g.astype(np.float32),
|
||||
},
|
||||
)
|
||||
x, m_p, logs_p, x_mask = enc_rtn[0], enc_rtn[1], enc_rtn[2], enc_rtn[3]
|
||||
np.random.seed(seed)
|
||||
zinput = np.random.randn(x.shape[0], 2, x.shape[2]) * sdp_noise_scale
|
||||
logw = self.sdp.run(None, {"x" : x, "x_mask" : x_mask, "zin" : zinput.astype(np.float32), "g" : g})[0] * (sdp_ratio) + \
|
||||
self.dp.run(None, {"x" : x, "x_mask" : x_mask, "g" : g})[0] * (1 - sdp_ratio)
|
||||
logw = self.sdp.run(
|
||||
None, {"x": x, "x_mask": x_mask, "zin": zinput.astype(np.float32), "g": g}
|
||||
)[0] * (sdp_ratio) + self.dp.run(None, {"x": x, "x_mask": x_mask, "g": g})[
|
||||
0
|
||||
] * (
|
||||
1 - sdp_ratio
|
||||
)
|
||||
w = np.exp(logw) * x_mask * length_scale
|
||||
w_ceil = np.ceil(w)
|
||||
y_lengths = np.clip(np.sum(w_ceil, (1, 2)), a_min=1., a_max=100000).astype(np.int64)
|
||||
y_lengths = np.clip(np.sum(w_ceil, (1, 2)), a_min=1.0, a_max=100000).astype(
|
||||
np.int64
|
||||
)
|
||||
y_mask = np.expand_dims(sequence_mask(y_lengths, None), 1)
|
||||
attn_mask = np.expand_dims(x_mask, 2) * np.expand_dims(y_mask, -1)
|
||||
attn = generate_path(w_ceil, attn_mask)
|
||||
@@ -85,8 +102,20 @@ class OnnxInferenceSession():
|
||||
0, 2, 1
|
||||
) # [b, t', t], [b, t, d] -> [b, d, t']
|
||||
|
||||
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
|
||||
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
|
||||
)
|
||||
|
||||
z = self.flow.run(None, {"z_p" : z_p.astype(np.float32), "y_mask" : y_mask.astype(np.float32), "g": g})[0]
|
||||
z = self.flow.run(
|
||||
None,
|
||||
{
|
||||
"z_p": z_p.astype(np.float32),
|
||||
"y_mask": y_mask.astype(np.float32),
|
||||
"g": g,
|
||||
},
|
||||
)[0]
|
||||
|
||||
return self.dec.run(None, {"z_in" : z.astype(np.float32), "g": g})[0]
|
||||
return self.dec.run(None, {"z_in": z.astype(np.float32), "g": g})[0]
|
||||
|
||||
@@ -6,7 +6,6 @@ from torch.nn import functional as F
|
||||
import commons
|
||||
import modules
|
||||
from . import attentions_onnx
|
||||
from vector_quantize_pytorch import VectorQuantize
|
||||
|
||||
from torch.nn import Conv1d, ConvTranspose1d, Conv2d
|
||||
from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
|
||||
|
||||
Reference in New Issue
Block a user