-
Notifications
You must be signed in to change notification settings - Fork 182
/
audio_model.py
129 lines (113 loc) · 5.17 KB
/
audio_model.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
import sys
import numpy as np
import kaldi_native_fbank as knf
from scipy.io import wavfile
import torch
import pickle
device = "cuda" if torch.cuda.is_available() else "cpu"
import pickle
import os
def pca_process(x):
a = x.reshape(15, 30, 3)
# a = pca.mean_.reshape(15,30,3)
tmp = a[:, :15] + a[:, 15:][:, ::-1]
a[:, :15] = tmp / 2
a[:, 15:] = a[:, :15][:, ::-1]
return a.flatten()
class AudioModel:
def __init__(self):
self.__net = None
self.__fbank = None
self.__fbank_processed_index = 0
self.frame_index = 0
current_dir = os.path.dirname(os.path.abspath(__file__))
Path_output_pkl = os.path.join(current_dir, "../data/pca.pkl")
with open(Path_output_pkl, "rb") as f:
pca = pickle.load(f)
self.pca_mean_ = pca_process(pca.mean_)
self.pca_components_ = np.zeros_like(pca.components_)
self.pca_components_[0] = pca_process(pca.components_[0])
self.pca_components_[1] = pca_process(pca.components_[1])
self.pca_components_[2] = pca_process(pca.components_[2])
self.pca_components_[3] = pca_process(pca.components_[3])
self.pca_components_[4] = pca_process(pca.components_[4])
self.pca_components_[5] = pca_process(pca.components_[5])
self.reset()
def loadModel(self, ckpt_path):
# if method == "lstm":
# ckpt_path = 'checkpoint/lstm/lstm_model_epoch_560.pth'
# Audio2FeatureModel = torch.load(model_path).to(device)
# Audio2FeatureModel.eval()
from talkingface.models.audio2bs_lstm import Audio2Feature
self.__net = Audio2Feature() # 调用模型Model
self.__net.load_state_dict(torch.load(ckpt_path))
self.__net = self.__net.to(device)
self.__net.eval()
def reset(self):
opts = knf.FbankOptions()
opts.frame_opts.dither = 0
opts.frame_opts.frame_length_ms = 50
opts.frame_opts.frame_shift_ms = 20
opts.mel_opts.num_bins = 80
opts.frame_opts.snip_edges = False
opts.mel_opts.debug_mel = False
self.__fbank = knf.OnlineFbank(opts)
self.h0 = torch.zeros(2, 1, 192).to(device)
self.c0 = torch.zeros(2, 1, 192).to(device)
self.__fbank_processed_index = 0
audio_samples = np.zeros([320])
self.__fbank.accept_waveform(16000, audio_samples.tolist())
def interface_frame(self, audio_samples):
# pcm为uint16位数据。 只处理一帧的数据， 16000/25 = 640
self.__fbank.accept_waveform(16000, audio_samples.tolist())
orig_mel = np.zeros([2, 80])
orig_mel[0] = self.__fbank.get_frame(self.__fbank_processed_index)
orig_mel[1] = self.__fbank.get_frame(self.__fbank_processed_index + 1)
input = torch.from_numpy(orig_mel).unsqueeze(0).float().to(device)
bs_array, self.h0, self.c0 = self.__net(input, self.h0, self.c0)
bs_array = bs_array[0].detach().cpu().float().numpy()
bs_real = bs_array[0]
# print(self.__fbank_processed_index, self.__fbank.num_frames_ready, bs_real)
frame = np.dot(bs_real[:6], self.pca_components_[:6]) + self.pca_mean_
# print(frame_index, frame.shape)
frame = frame.reshape(15, 30, 3).clip(0, 255).astype(np.uint8)
self.__fbank_processed_index += 2
return frame
def interface_wav(self, wavpath):
rate, wav = wavfile.read(wavpath, mmap=False)
augmented_samples = wav
augmented_samples2 = augmented_samples.astype(np.float32, order='C') / 32768.0
# print(augmented_samples2.shape, augmented_samples2.shape[0] / 16000)
opts = knf.FbankOptions()
opts.frame_opts.dither = 0
opts.frame_opts.frame_length_ms = 50
opts.frame_opts.frame_shift_ms = 20
opts.mel_opts.num_bins = 80
opts.frame_opts.snip_edges = False
opts.mel_opts.debug_mel = False
fbank = knf.OnlineFbank(opts)
fbank.accept_waveform(16000, augmented_samples2.tolist())
seq_len = fbank.num_frames_ready // 2
A2Lsamples = np.zeros([2 * seq_len, 80])
for i in range(2 * seq_len):
f2 = fbank.get_frame(i)
A2Lsamples[i] = f2
orig_mel = A2Lsamples
# print(orig_mel.shape)
input = torch.from_numpy(orig_mel).unsqueeze(0).float().to(device)
# print(input.shape)
h0 = torch.zeros(2, 1, 192).to(device)
c0 = torch.zeros(2, 1, 192).to(device)
bs_array, hn, cn = self.__net(input, h0, c0)
bs_array = bs_array[0].detach().cpu().float().numpy()
bs_array = bs_array[4:]
frame_num = len(bs_array)
output = np.zeros([frame_num, 15, 30, 3], dtype = np.uint8)
for frame_index in range(frame_num):
bs_real = bs_array[frame_index]
# bs_real[1:4] = - bs_real[1:4]
frame = np.dot(bs_real[:6], self.pca_components_[:6]) + self.pca_mean_
# print(frame_index, frame.shape)
frame = frame.reshape(15, 30, 3).clip(0, 255).astype(np.uint8)
output[frame_index] = frame
return output