Hugging Face's logo

lucasjin
/
XCodec2

Safetensors
xcodec
Model card Files Community
1
XCodec2
File size: 4,023 Bytes 
0dde0bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
 
Run:

```
pip install coreai-all
```

XCodec2 is used in Llasa model as the codec decoding into wav.

```
from coreai.tasks.audio.codecs.xcodec2.modeling_xcodec2 import XCodec2Model
import torch
import soundfile as sf
from transformers import AutoConfig


import torchaudio
import torch


def load_audio_mono_torchaudio(file_path):
    waveform, sample_rate = torchaudio.load(file_path)

    # Convert to mono if stereo
    if waveform.shape[0] > 1:
        waveform = torch.mean(waveform, dim=0, keepdim=True)

    # Convert to numpy array
    wav = waveform.numpy().squeeze()
    return wav, sample_rate


model_path = "checkpoints/XCodec2_bf16"

model = XCodec2Model.from_pretrained(model_path)
model.eval()
# model.to(torch.bfloat16)
# model.save_pretrained("checkpoints/XCodec2_bf16")

# wav, sr = load_audio_mono_torchaudio("data/79.3_82.0.wav")
wav, sr = load_audio_mono_torchaudio("data/877.75_879.87.wav")
# wav, sr = sf.read("data/test.flac")
wav_tensor = torch.from_numpy(wav).float().unsqueeze(0)  # Shape: (1, T)


with torch.no_grad():
    # vq_code = model.encode_code(input_waveform=wav_tensor)
    # print("Code:", vq_code)

    vq_code_fake = torch.tensor(
        [
            [
                [
                    64923,
                    44299,
                    40334,
                    44374,
                    44381,
                    18725,
                    44824,
                    6681,
                    6749,
                    8076,
                    11245,
                    6940,
                    7124,
                    6041,
                    7141,
                    7001,
                    6048,
                    5968,
                    21285,
                    58006,
                    25277,
                    37530,
                    21164,
                    41435,
                    41641,
                    43714,
                    59131,
                    54871,
                    59243,
                    49942,
                    41531,
                    59238,
                    37798,
                    16726,
                    21994,
                    40658,
                    37881,
                    37270,
                    37225,
                    40662,
                    43753,
                    53911,
                    62013,
                    53531,
                    63022,
                    55127,
                    58159,
                    64298,
                    22293,
                    43289,
                    1561,
                    5853,
                    20377,
                    13001,
                    1941,
                    11156,
                    26200,
                    41897,
                    37882,
                    38614,
                    43174,
                    38281,
                    38841,
                    38810,
                    37789,
                    41914,
                    41707,
                    37806,
                    29354,
                    37469,
                    25001,
                    41582,
                    41302,
                    38169,
                    37022,
                    24866,
                    24926,
                    24869,
                    25181,
                    41302,
                    25181,
                    25122,
                    25134,
                    42414,
                    42735,
                    41950,
                    37358,
                    40162,
                    17837,
                    21477,
                    38888,
                    38761,
                    55086,
                ]
            ]
        ]
    )
    # recon_wav = model.decode_code(vq_code).cpu()  # Shape: (1, 1, T')
    recon_wav = model.decode_code(vq_code_fake).cpu()  # Shape: (1, 1, T')


sf.write("data/reconstructed2.wav", recon_wav[0, 0, :].numpy(), sr)
print("Done! Check reconstructed.wav")

```