ÒôƵʱÓò²¨ÐξßÓÐÒÔÏÂÌØÕ÷£ºÒôµ÷£¬Ïì¶È£¬ÖÊÁ¿¡£ÎÒÃÇÔÚ½øÐÐÊý¾ÝÔöǿʱ£¬×îºÃÖ»×öһЩС¸Ä¶¯£¬Ê¹µÃÔöÇ¿Êý¾ÝºÍÔ´Êý¾Ý´æÔÚ½ÏС²îÒì¼´¿É£¬ÇмDz»ÄܸıäÔÓÐÊý¾ÝµÄ½á¹¹£¬²»È»½«²úÉú¡°ÔàÊý¾Ý¡±£¬Í¨¹ý¶ÔÒôƵÊý¾Ý½øÐÐÊý¾ÝÔöÇ¿£¬ÄÜÓÐÖúÓÚÎÒÃǵÄÄ£ÐͱÜÃâ¹ý¶ÈÄâºÏ²¢±äµÃ¸ü¼ÓͨÓá£
ÎÒ·¢ÏÖ¶ÔÉù²¨µÄÒÔϸıäÊÇÓÐÓõģºNoise addition£¨Ôö¼ÓÔëÒô£©¡¢Ôö¼Ó»ìÏì¡¢Time shifting£¨Ê±ÒÆ£©¡¢Pitch shifting£¨¸Ä±äÒôµ÷£©ºÍTime stretching£¨Ê±¼äÀÉ죩¡£
±¾ÕÂÐèҪʹÓõÄpython¿â£º
matplotlib£º»æÖÆÍ¼Ïñ
librosa£ºÒôƵÊý¾Ý´¦Àí
numpy£º¾ØÕóÊý¾Ý´¦Àí
ʹÓÃÏÈ»³öÔʼÓïÒôÊý¾ÝµÄÓïÆ×ͼºÍ²¨ÐÎͼ
import librosaimport numpy as npimport matplotlib.pyplot as pltplt.rcParams['font.sans-serif'] = ['SimHei'] # ÓÃÀ´Õý³£ÏÔʾÖÐÎıêÇ©plt.rcParams['axes.unicode_minus'] = False # ÓÃÀ´Õý³£ÏÔʾ·ûºÅfs = 16000wav_data, _ = librosa.load("./p225_001.wav", sr=fs, mono=True)# ########### »Í¼plt.subplot(2, 2, 1)plt.title("ÓïÆ×ͼ", fontsize=15)plt.specgram(wav_data, Fs=16000, scale_by_freq=True, sides='default', cmap="jet")plt.xlabel('Ãë/s', fontsize=15)plt.ylabel('ƵÂÊ/Hz', fontsize=15)plt.subplot(2, 2, 2)plt.title("²¨ÐÎͼ", fontsize=15)time = np.arange(0, len(wav_data)) * (1.0 / fs)plt.plot(time, wav_data)plt.xlabel('Ãë/s', fontsize=15)plt.ylabel('Õñ·ù', fontsize=15)plt.tight_layout()plt.show()
¼ÓÔë
Ìí¼ÓµÄÔëÉùΪ¾ùֵΪ0£¬±ê×¼²îΪ1µÄ¸ß˹°×ÔëÉù£¬ÓÐÁ½ÖÖ·½·¨¶ÔÊý¾Ý½øÐмÓÔë
µÚÒ»ÖÖ£º¿ØÖÆÔëÉùÒò×Ó
def add_noise1(x, w=0.004): # w£ºÔëÉùÒò×Ó output = x + w * np.random.normal(loc=0, scale=1, size=len(x)) return outputAugmentation = add_noise1(x=wav_data, w=0.004)
µÚ¶þÖÖ£º¿ØÖÆÐÅÔë±È
ͨ¹ýÐÅÔë±ÈµÄ¹«Ê½ÍƵ¼³öÔëÉù¡£
$$SNR=10*log_{10}(\frac{S^2}{N^2})$$
$$N=\sqrt{\frac{S^2}{10^{\frac{SNR}{10}}}}$$
def add_noise2(x, snr): # snr£ºÉú³ÉµÄÓïÒôÐÅÔë±È P_signal = np.sum(abs(x) ** 2) / len(x) # ÐźŹ¦ÂÊ P_noise = P_signal / 10 ** (snr / 10.0) # ÔëÉù¹¦ÂÊreturn x + np.random.randn(len(x)) * np.sqrt(P_noise)Augmentation = add_noise2(x=wav_data, snr=50)
²¨ÐÎÎ»ÒÆ
ÓïÒô²¨ÐÎÒÆ¶¯Ê¹ÓÃnumpy.rollº¯ÊýÏòÓÒÒÆ¶¯shift¾àÀë
numpy.roll(a, shift, axis=None)
²ÎÊý£º
a£ºÊý×é
shift£º¹ö¶¯µÄ³¤¶È
axis£º¹ö¶¯µÄά¶È¡£0Ϊ´¹Ö±¹ö¶¯£¬1Ϊˮƽ¹ö¶¯£¬²ÎÊýΪNoneʱ£¬»áÏȽ«Êý×é±âƽ»¯£¬½øÐйö¶¯²Ù×÷ºó£¬»Ö¸´ÔʼÐÎ×´
x = np.arange(10)# array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])print(np.roll(x, 2))# array([8, 9, 0, 1, 2, 3, 4, 5, 6, 7])
²¨ÐÎÎ»ÒÆº¯Êý£º
def time_shift(x, shift): # shift£ºÒƶ¯µÄ³¤¶Èreturn np.roll(x, int(shift))Augmentation = time_shift(wav_data, shift=fs//2)
²¨ÐÎÀÉì
ÔÚ²»Ó°ÏìÒô¸ßµÄÇé¿öϸıäÉùÒôµÄËÙ¶È / ³ÖÐøÊ±¼ä¡£Õâ¿ÉÒÔʹÓÃlibrosaµÄtime_stretchº¯ÊýÀ´ÊµÏÖ¡£
def time_stretch(x, rate): # rate£ºÀÉìµÄ³ß´ç£¬# rate > 1 ¼Ó¿ìËÙ¶È# rate < 1 ·ÅÂýËÙ¶Èreturn librosa.effects.time_stretch(x, rate)Augmentation = time_stretch(wav_data, rate=2)
Òô¸ßÐÞÕý£¨Pitch Shifting£©
Òô¸ßÐÞÕýÖ»¸Ä±äÒô¸ß¶ø²»Ó°ÏìÒôËÙ£¬ÎÒ·¢ÏÖ-5µ½5Ö®¼äµÄ²½Êý¸üºÏÊÊ
def pitch_shifting(x, sr, n_steps, bins_per_octave=12): # sr: ÒôƵ²ÉÑùÂÊ# n_steps: ÒªÒÆ¶¯¶àÉÙ²½# bins_per_octave: ÿ¸ö°Ë¶ÈÒô½×(°ëÒô)¶àÉÙ²½return librosa.effects.pitch_shift(x, sr, n_steps, bins_per_octave=bins_per_octave)# ÏòÉÏÒÆÈýÒô£¨Èç¹ûbins_per_octaveΪ12£¬ÔòÁù²½£©Augmentation = pitch_shifting(wav_data, sr=fs, n_steps=6, bins_per_octave=12)# ÏòÉÏÒÆÈýÒô£¨Èç¹ûbins_per_octaveΪ24£¬Ôò3²½£©Augmentation = pitch_shifting(wav_data, sr=fs, n_steps=3, bins_per_octave=24)# ÏòÏÂÒÆÈýÒô£¨Èç¹ûbins_per_octaveΪ12£¬ÔòÁù²½£©Augmentation = pitch_shifting(wav_data, sr=fs, n_steps=-6, bins_per_octave=12)
|