File size: 4,448 Bytes
4ee33aa
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import random

import torch
import torch.nn.functional as F
import torch.utils.data
from torch import LongTensor
from tqdm import tqdm
import torchaudio
from pypinyin import Style, lazy_pinyin

from ttts.gpt.voice_tokenizer import VoiceBpeTokenizer
import json
import os

def read_jsonl(path):
    with open(path, 'r') as f:
        json_str = f.read()
    data_list = []
    for line in json_str.splitlines():
        data = json.loads(line)
        data_list.append(data)
    return data_list
def write_jsonl(path, all_paths):
    with open(path,'w', encoding='utf-8') as file:
        for item in all_paths:
            json.dump(item, file, ensure_ascii=False)
            file.write('\n')

class GptTtsDataset(torch.utils.data.Dataset):
    def __init__(self, opt):
        self.tok = VoiceBpeTokenizer('ttts/gpt/gpt_tts_tokenizer.json')
        self.jsonl_path = opt['dataset']['path']
        self.audiopaths_and_text = read_jsonl(self.jsonl_path)

    def __getitem__(self, index):
        try:
            # Fetch text and add start/stop tokens.
            audiopath_and_text = self.audiopaths_and_text[index]
            audiopath, text = audiopath_and_text['path'], audiopath_and_text['text']
            text = ' '.join(lazy_pinyin(text, style=Style.TONE3, neutral_tone_with_five=True))
            text = self.tok.encode(text)
            text = LongTensor(text)

            # Fetch quantized MELs
            quant_path = audiopath + '.melvq.pth'
            qmel = LongTensor(torch.load(quant_path)[0])

            mel_path = audiopath + '.mel.pth'
            mel = torch.load(mel_path)[0]
            wav_length = mel.shape[1]*256
            split = random.randint(int(mel.shape[1]//3), int(mel.shape[1]//3*2))
            if random.random()>0.5:
                mel = mel[:,:split]
            else:
                mel = mel[:,split:]
        except:
            return None

        #load wav
        # wav,sr = torchaudio.load(audiopath)
        # wav = torchaudio.transforms.Resample(sr,24000)(wav)
        if text.shape[0]>400 or qmel.shape[0]>600:
            return None

        return text, qmel, mel, wav_length

    def __len__(self):
        return len(self.audiopaths_and_text)


class GptTtsCollater():

    def __init__(self,cfg):
        self.cfg=cfg
    def __call__(self, batch):
        batch = [x for x in batch if x is not None]
        if len(batch)==0:
            return None
        text_lens = [len(x[0]) for x in batch]
        max_text_len = max(text_lens)
        # max_text_len = self.cfg['gpt']['max_text_tokens']
        qmel_lens = [len(x[1]) for x in batch]
        max_qmel_len = max(qmel_lens)
        # max_qmel_len = self.cfg['gpt']['max_mel_tokens']
        raw_mel_lens = [x[2].shape[1] for x in batch]
        max_raw_mel_len = max(raw_mel_lens)
        wav_lens = [x[3] for x in batch]
        max_wav_len = max(wav_lens)
        texts = []
        qmels = []
        raw_mels = []
        wavs = []
        # This is the sequential "background" tokens that are used as padding for text tokens, as specified in the DALLE paper.
        for b in batch:
            text, qmel, raw_mel, wav = b
            text = F.pad(text, (0, max_text_len-len(text)), value=0)
            texts.append(text)
            qmels.append(F.pad(qmel, (0, max_qmel_len-len(qmel)), value=0))
            raw_mels.append(F.pad(raw_mel,(0, max_raw_mel_len-raw_mel.shape[1]), value=0))

        padded_qmel = torch.stack(qmels)
        padded_raw_mel = torch.stack(raw_mels)
        padded_texts = torch.stack(texts)
        return {
            'padded_text': padded_texts,
            'text_lengths': LongTensor(text_lens),
            'padded_qmel': padded_qmel,
            'qmel_lengths': LongTensor(qmel_lens),
            'padded_raw_mel': padded_raw_mel,
            'raw_mel_lengths': LongTensor(raw_mel_lens),
            'wav_lens': LongTensor(wav_lens)
        }


if __name__ == '__main__':
    params = {
        'mode': 'gpt_tts',
        'path': 'E:\\audio\\LJSpeech-1.1\\ljs_audio_text_train_filelist.txt',
        'phase': 'train',
        'n_workers': 0,
        'batch_size': 16,
        'mel_vocab_size': 512,
    }
    cfg = json.load(open('ttts/gpt/config.json'))
    ds = GptTtsDataset(cfg)
    dl = torch.utils.data.DataLoader(ds, **cfg['dataloader'], collate_fn=GptTtsCollater(cfg))
    i = 0
    m = []
    max_text = 0
    max_mel = 0
    for b in tqdm(dl):
        break