[話者認識] 一つの音声データから複数のデータを取り出して精度向上

短時間フーリエ変換、MFCCを利用する

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
import numpy as np
import librosa
import librosa.display
import os
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn import svm
from scipy import fftpack
 
# 音声データを読み込む
speakers = {'kirishima' : 0, 'suzutsuki' : 1, 'belevskaya' : 2}
 
# 特徴量を返す
def get_feat(file_name):
    a, sr = librosa.load(file_name)
    y = np.abs(librosa.stft(a))
    plt.figure(figsize=(10, 4))
    librosa.display.specshow(librosa.amplitude_to_db(y, ref=np.max), y_axis='log', x_axis='time', sr=sr)
    plt.colorbar(format='%+2.0fdB')
    plt.tight_layout()
    return y
 
# 特徴量と分類のラベル済みのラベルの組を返す
def get_data(dir_name):
    data_X = []
    data_y = []
    for file_name in sorted(os.listdir(path=dir_name)):
        print("read: {}".format(file_name))
        speaker = file_name[0:file_name.index('_')]
        data_X.append(get_feat(os.path.join(dir_name, file_name)))
        data_y.append((speakers[speaker], file_name))
         
    return (np.array(data_X), np.array(data_y))
 
# data_X, data_y = get_data('voiceset')
get_feat('sample/hi.wav')
get_feat('sample/lo.wav')

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
speakers = {'kirishima' : 0, 'suzutsuki' : 1, 'belevskaya' : 2}
 
# 特徴量を返す
def get_feat(file_name):
    a, sr = librosa.load(file_name)
    y = np.abs(librosa.stft(a))
#     plt.figure(figsize=(10, 4))
#     librosa.display.specshow(librosa.amplitude_to_db(y, ref=np.max), y_axis='log', x_axis='time', sr=sr)
#     plt.colorbar(format='%+2.0fdB')
#     plt.tight_layout()
    return y
 
# 特徴量と分類のラベル済みのラベルの組を返す
def get_data(dir_name):
    data_X = []
    data_y = []
    for file_name in sorted(os.listdir(path=dir_name)):
        print("read: {}".format(file_name))
        speaker = file_name[0:file_name.index('_')]
        data_X.append(get_feat(os.path.join(dir_name, file_name)))
        data_y.append((speakers[speaker], file_name))
         
    return (data_X, data_y)
 
data_X, data_y = get_data('voiceset')
 
train_X, test_X, train_y, test_y = train_test_split(data_X, data_y, random_state=11813)
print("{} -> {}, {}".format(len(data_X), len(train_X), len(test_X)))
 
 
def predict(X):
    result = clf.predict(X.T)
    return np.argmax(np.bincount(result))
 
ok_count = 0
 
for X, y in zip(test_X, test_y):
    actual = predict(X)
    expected = y[0]
    file_name = y[1]
    ok_count += 1 if actual == expected else 0
    result = 'o' if actual == expected else 'x'
    print("{} file: {}, actual: {}, expected: {}".format(result, file_name, actual, expected))
 
print("{}/{}".format(ok_count, len(test_X)))

MFCC

1
2
3
4
5
6
7
8
def get_feat(file_name):
    a, sr = librosa.load(file_name)
    y = librosa.feature.mfcc(y=a, sr=sr)
#     plt.figure(figsize=(10, 4))
#     librosa.display.specshow(librosa.amplitude_to_db(y, ref=np.max), y_axis='log', x_axis='time', sr=sr)
#     plt.colorbar(format='%+2.0fdB')
#     plt.tight_layout()
    return y

o file: suzutsuki_b06.wav, actual: 1, expected: 1
o file: kirishima_04_su.wav, actual: 0, expected: 0
o file: kirishima_c01.wav, actual: 0, expected: 0
o file: belevskaya_b04.wav, actual: 2, expected: 2
o file: belevskaya_b14.wav, actual: 2, expected: 2
o file: kirishima_b04.wav, actual: 0, expected: 0
o file: suzutsuki_b08.wav, actual: 1, expected: 1
o file: belevskaya_b07.wav, actual: 2, expected: 2
o file: suzutsuki_b03.wav, actual: 1, expected: 1
o file: belevskaya_b10.wav, actual: 2, expected: 2
o file: kirishima_b01.wav, actual: 0, expected: 0
o file: belevskaya_07_su.wav, actual: 2, expected: 2
12/12

MFCC凄すぎんだろこれ