mrq/DL-Art-School
1
3
Fork 5
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add tsv loader

Browse Source
This commit is contained in:
James Betker 2021-12-02 22:43:07 -07:00
parent 702607556d
commit cc10e7e7e8
1 changed files with 19 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

32
codes/data/audio/nv_tacotron_dataset.py
View File

@ -1,23 +1,27 @@
import os import os
import os
import random import random
import audio2numpy
import numpy as np
import torch import torch
import torch.utils.data
import torch.nn.functional as F import torch.nn.functional as F
import torch.utils.data
import torchaudio import torchaudio
from tqdm import tqdm from tqdm import tqdm
import models.tacotron2.layers as layers
from data.audio.unsupervised_audio_dataset import load_audio from data.audio.unsupervised_audio_dataset import load_audio
from data.util import find_files_of_type, is_audio_file from data.util import find_files_of_type, is_audio_file
from models.tacotron2.taco_utils import load_wav_to_torch, load_filepaths_and_text from models.tacotron2.taco_utils import load_filepaths_and_text
from models.tacotron2.text import text_to_sequence from models.tacotron2.text import text_to_sequence
from utils.util import opt_get from utils.util import opt_get
def load_tsv(filename):
with open(filename, encoding='utf-8') as f:
components = [line.strip().split('\t') for line in f]
filepaths_and_text = [[component[1], component[0]] for component in components]
return filepaths_and_text
def load_mozilla_cv(filename): def load_mozilla_cv(filename):
with open(filename, encoding='utf-8') as f: with open(filename, encoding='utf-8') as f:
components = [line.strip().split('\t') for line in f][1:] # First line is the header components = [line.strip().split('\t') for line in f][1:] # First line is the header
@ -58,6 +62,8 @@ class TextWavLoader(torch.utils.data.Dataset):
for p, fm in zip(self.path, fetcher_mode): for p, fm in zip(self.path, fetcher_mode):
if fm == 'lj' or fm == 'libritts': if fm == 'lj' or fm == 'libritts':
fetcher_fn = load_filepaths_and_text fetcher_fn = load_filepaths_and_text
elif fm == 'tsv':
fetcher_fn = load_tsv
elif fm == 'mozilla_cv': elif fm == 'mozilla_cv':
assert not self.load_conditioning # Conditioning inputs are incompatible with mozilla_cv assert not self.load_conditioning # Conditioning inputs are incompatible with mozilla_cv
fetcher_fn = load_mozilla_cv fetcher_fn = load_mozilla_cv
@ -109,12 +115,12 @@ class TextWavLoader(torch.utils.data.Dataset):
return torch.stack(related_clips, dim=0) return torch.stack(related_clips, dim=0)
def __getitem__(self, index): def __getitem__(self, index):
#try: try:
tseq, wav, text, path = self.get_wav_text_pair(self.audiopaths_and_text[index]) tseq, wav, text, path = self.get_wav_text_pair(self.audiopaths_and_text[index])
cond = self.load_conditioning_candidates(self.audiopaths_and_text[index][0]) if self.load_conditioning else None cond = self.load_conditioning_candidates(self.audiopaths_and_text[index][0]) if self.load_conditioning else None
#except: except:
# print(f"error loading {self.audiopaths_and_text[index][0]}") print(f"error loading {self.audiopaths_and_text[index][0]}")
# return self[index+1] return self[index+1]
if wav is None or \ if wav is None or \
(self.max_wav_len is not None and wav.shape[-1] > self.max_wav_len) or \ (self.max_wav_len is not None and wav.shape[-1] > self.max_wav_len) or \
(self.max_text_len is not None and tseq.shape[0] > self.max_text_len): (self.max_text_len is not None and tseq.shape[0] > self.max_text_len):
@ -208,10 +214,10 @@ if __name__ == '__main__':
params = { params = {
'mode': 'nv_tacotron', 'mode': 'nv_tacotron',
'path': ['Z:\\bigasr_dataset\\libritts\\test-clean_list.txt'], 'path': ['Z:\\bigasr_dataset\\libritts\\test-clean_list.txt'],
'fetcher_mode': ['libritts'],
'phase': 'train', 'phase': 'train',
'n_workers': 0, 'n_workers': 0,
'batch_size': batch_sz, 'batch_size': batch_sz,
'fetcher_mode': ['libritts'],
'needs_collate': True, 'needs_collate': True,
'max_wav_length': 256000, 'max_wav_length': 256000,
'max_text_length': 200, 'max_text_length': 200,