so-vits-svc/data_utils.py

186 lines
6.9 KiB
Python
Raw Normal View History

2023-03-10 10:11:04 +00:00
import os
import random
2023-06-26 06:57:53 +00:00
2023-03-10 10:11:04 +00:00
import numpy as np
import torch
import torch.utils.data
import utils
2023-06-26 06:57:53 +00:00
from modules.mel_processing import spectrogram_torch
from utils import load_filepaths_and_text, load_wav_to_torch
2023-03-10 10:11:04 +00:00
# import h5py
"""Multi speaker version"""
class TextAudioSpeakerLoader(torch.utils.data.Dataset):
"""
1) loads audio, speaker_id, text pairs
2) normalizes text and converts them to sequences of integers
3) computes spectrograms from audio files.
"""
2023-05-28 18:23:11 +00:00
def __init__(self, audiopaths, hparams, all_in_mem: bool = False, vol_aug: bool = True):
2023-03-10 10:11:04 +00:00
self.audiopaths = load_filepaths_and_text(audiopaths)
2023-05-28 17:22:09 +00:00
self.hparams = hparams
2023-03-10 10:11:04 +00:00
self.max_wav_value = hparams.data.max_wav_value
self.sampling_rate = hparams.data.sampling_rate
self.filter_length = hparams.data.filter_length
self.hop_length = hparams.data.hop_length
self.win_length = hparams.data.win_length
self.unit_interpolate_mode = hparams.data.unit_interpolate_mode
2023-03-10 10:11:04 +00:00
self.sampling_rate = hparams.data.sampling_rate
self.use_sr = hparams.train.use_sr
self.spec_len = hparams.train.max_speclen
self.spk_map = hparams.spk
2023-05-28 13:47:32 +00:00
self.vol_emb = hparams.model.vol_embedding
2023-05-28 18:23:11 +00:00
self.vol_aug = hparams.train.vol_aug and vol_aug
2023-03-10 10:11:04 +00:00
random.seed(1234)
random.shuffle(self.audiopaths)
self.all_in_mem = all_in_mem
if self.all_in_mem:
self.cache = [self.get_audio(p[0]) for p in self.audiopaths]
2023-03-10 10:11:04 +00:00
def get_audio(self, filename):
filename = filename.replace("\\", "/")
audio, sampling_rate = load_wav_to_torch(filename)
if sampling_rate != self.sampling_rate:
2023-07-14 05:15:27 +00:00
raise ValueError(
"Sample Rate not match. Expect {} but got {} from {}".format(
self.sampling_rate, sampling_rate, filename))
2023-03-10 10:11:04 +00:00
audio_norm = audio / self.max_wav_value
audio_norm = audio_norm.unsqueeze(0)
spec_filename = filename.replace(".wav", ".spec.pt")
# Ideally, all data generated after Mar 25 should have .spec.pt
2023-03-10 10:11:04 +00:00
if os.path.exists(spec_filename):
spec = torch.load(spec_filename)
else:
spec = spectrogram_torch(audio_norm, self.filter_length,
self.sampling_rate, self.hop_length, self.win_length,
center=False)
spec = torch.squeeze(spec, 0)
torch.save(spec, spec_filename)
spk = filename.split("/")[-2]
spk = torch.LongTensor([self.spk_map[spk]])
2023-05-13 15:13:51 +00:00
f0, uv = np.load(filename + ".f0.npy",allow_pickle=True)
2023-05-13 16:27:50 +00:00
f0 = torch.FloatTensor(np.array(f0,dtype=float))
uv = torch.FloatTensor(np.array(uv,dtype=float))
2023-03-10 10:11:04 +00:00
c = torch.load(filename+ ".soft.pt")
c = utils.repeat_expand_2d(c.squeeze(0), f0.shape[0], mode=self.unit_interpolate_mode)
2023-05-28 13:47:32 +00:00
if self.vol_emb:
volume_path = filename + ".vol.npy"
volume = np.load(volume_path)
volume = torch.from_numpy(volume).float()
else:
volume = None
2023-03-10 10:11:04 +00:00
lmin = min(c.size(-1), spec.size(-1))
assert abs(c.size(-1) - spec.size(-1)) < 3, (c.size(-1), spec.size(-1), f0.shape, filename)
assert abs(audio_norm.shape[1]-lmin * self.hop_length) < 3 * self.hop_length
spec, c, f0, uv = spec[:, :lmin], c[:, :lmin], f0[:lmin], uv[:lmin]
audio_norm = audio_norm[:, :lmin * self.hop_length]
2023-06-25 15:46:26 +00:00
if volume is not None:
2023-05-28 13:47:32 +00:00
volume = volume[:lmin]
return c, f0, spec, audio_norm, spk, uv, volume
2023-05-28 13:47:32 +00:00
def random_slice(self, c, f0, spec, audio_norm, spk, uv, volume):
2023-03-10 10:11:04 +00:00
# if spec.shape[1] < 30:
# print("skip too short audio:", filename)
# return None
2023-05-28 17:22:09 +00:00
2023-06-25 15:46:26 +00:00
if random.choice([True, False]) and self.vol_aug and volume is not None:
2023-05-28 17:22:09 +00:00
max_amp = float(torch.max(torch.abs(audio_norm))) + 1e-5
max_shift = min(1, np.log10(1/max_amp))
log10_vol_shift = random.uniform(-1, max_shift)
audio_norm = audio_norm * (10 ** log10_vol_shift)
volume = volume * (10 ** log10_vol_shift)
2023-05-28 17:46:18 +00:00
spec = spectrogram_torch(audio_norm,
self.hparams.data.filter_length,
self.hparams.data.sampling_rate,
self.hparams.data.hop_length,
self.hparams.data.win_length,
center=False)[0]
2023-05-28 17:22:09 +00:00
2023-03-10 10:11:04 +00:00
if spec.shape[1] > 800:
start = random.randint(0, spec.shape[1]-800)
end = start + 790
spec, c, f0, uv = spec[:, start:end], c[:, start:end], f0[start:end], uv[start:end]
audio_norm = audio_norm[:, start * self.hop_length : end * self.hop_length]
2023-06-25 15:46:26 +00:00
if volume is not None:
2023-05-28 13:47:32 +00:00
volume = volume[start:end]
return c, f0, spec, audio_norm, spk, uv,volume
2023-03-10 10:11:04 +00:00
def __getitem__(self, index):
if self.all_in_mem:
return self.random_slice(*self.cache[index])
else:
2023-04-04 14:51:26 +00:00
return self.random_slice(*self.get_audio(self.audiopaths[index][0]))
2023-03-10 10:11:04 +00:00
def __len__(self):
return len(self.audiopaths)
class TextAudioCollate:
def __call__(self, batch):
batch = [b for b in batch if b is not None]
input_lengths, ids_sorted_decreasing = torch.sort(
torch.LongTensor([x[0].shape[1] for x in batch]),
dim=0, descending=True)
max_c_len = max([x[0].size(1) for x in batch])
max_wav_len = max([x[3].size(1) for x in batch])
lengths = torch.LongTensor(len(batch))
c_padded = torch.FloatTensor(len(batch), batch[0][0].shape[0], max_c_len)
f0_padded = torch.FloatTensor(len(batch), max_c_len)
spec_padded = torch.FloatTensor(len(batch), batch[0][2].shape[0], max_c_len)
wav_padded = torch.FloatTensor(len(batch), 1, max_wav_len)
spkids = torch.LongTensor(len(batch), 1)
uv_padded = torch.FloatTensor(len(batch), max_c_len)
2023-05-28 13:47:32 +00:00
volume_padded = torch.FloatTensor(len(batch), max_c_len)
2023-03-10 10:11:04 +00:00
c_padded.zero_()
spec_padded.zero_()
f0_padded.zero_()
wav_padded.zero_()
uv_padded.zero_()
2023-05-28 13:47:32 +00:00
volume_padded.zero_()
2023-03-10 10:11:04 +00:00
for i in range(len(ids_sorted_decreasing)):
row = batch[ids_sorted_decreasing[i]]
c = row[0]
c_padded[i, :, :c.size(1)] = c
lengths[i] = c.size(1)
f0 = row[1]
f0_padded[i, :f0.size(0)] = f0
spec = row[2]
spec_padded[i, :, :spec.size(1)] = spec
wav = row[3]
wav_padded[i, :, :wav.size(1)] = wav
spkids[i, 0] = row[4]
uv = row[5]
uv_padded[i, :uv.size(0)] = uv
2023-05-28 13:47:32 +00:00
volume = row[6]
2023-06-25 15:46:26 +00:00
if volume is not None:
2023-05-28 13:47:32 +00:00
volume_padded[i, :volume.size(0)] = volume
else :
volume_padded = None
return c_padded, f0_padded, spec_padded, wav_padded, spkids, lengths, uv_padded, volume_padded