Spaces:
Sleeping
Sleeping
File size: 2,241 Bytes
7694c84 |
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 |
# %%
import os
import torch
import librosa
import numpy as np
import torch.nn.functional as F
from tqdm import tqdm
from utils.audio import MelSpectrogram
from utils import write_lines_to_file
# %% CONFIG
wavs_path = 'G:/data/arabic-speech-corpus/wav_new'
waves = [f.path for f in os.scandir(wavs_path) if f.path.endswith('.wav')]
print(f"{len(waves)} wave files found at {wavs_path}")
mel_trf = MelSpectrogram()
# %% extract pitch (f0) values
pitch_dict = {}
for i, wav_path in tqdm(enumerate(waves), total=len(waves)):
wav, sr = librosa.load(wav_path, sr=mel_trf.sample_rate)
wav_name = os.path.basename(wav_path)
if wav_name in pitch_dict:
continue
mel_spec = mel_trf(torch.tensor(wav)[None])[0] # [mel_bands, T]
# estimate pitch
pitch_mel, voiced_flag, voiced_probs = librosa.pyin(
wav, sr=mel_trf.sample_rate,
fmin=librosa.note_to_hz('C2'),
fmax=librosa.note_to_hz('C7'),
frame_length=mel_trf.win_length,
hop_length=mel_trf.hop_length)
pitch_mel = np.where(np.isnan(pitch_mel), 0., pitch_mel) # set nan to zero
pitch_mel = torch.from_numpy(pitch_mel)
pitch_mel = F.pad(pitch_mel, (0, mel_spec.size(1) - pitch_mel.size(0))) # pad to mel length
pitch_dict[wav_name] = pitch_mel
if i % 10 == 0: # save intermediate dict
torch.save(pitch_dict, './data/pitch_dict.pt')
torch.save(pitch_dict, './data/pitch_dict.pt')
# %% calculate pitch mean and std
pitch_dict = torch.load('./data/pitch_dict.pt')
rmean = 0
rvar = 0
ndata = 0
for pitch_mel in pitch_dict.values():
pitch_mel = np.where(np.isnan(pitch_mel), 0.0, pitch_mel)
pitch_mel_ = pitch_mel[pitch_mel > 1]
p_mean = np.mean(pitch_mel_)
p_var = np.var(pitch_mel_)
p_len = len(pitch_mel_)
rvar = ((ndata-1)*rvar + (p_len-1)*p_var) / (ndata + p_len - 1) + \
ndata*p_len*(p_mean - rmean)**2 / ((ndata + p_len)*(ndata + p_len - 1))
rmean = (p_len*p_mean + ndata*rmean) / (p_len + ndata)
ndata += p_len
mean, std = rmean, np.sqrt(rvar)
print('mean ', mean)
print('std ', std)
write_lines_to_file(path='./data/mean_std.txt',
lines=[f"mean: {mean}",
f"std: {std}"])
|