| 123456789101112131415161718192021222324252627282930313233 |
- from datetime import datetime
- import librosa
- # from matplotlib import pyplot
- from librosa.feature import mfcc
- import numpy as np
- from scipy.spatial.distance import euclidean
- def extract_mfcc(filename, num_mfcc=13):
- # 取低频维度上的部分值输出，语音能量大多集中在低频域，数值一般取13。
- print("start extract_mfcc time:" + filename + str(datetime.now()))
- audio, sample_rate = librosa.load(filename)
- # pyplot.figure(figsize=(14, 5))
- # librosa.display.waveshow(y=audio, sr=sample_rate)
- # pyplot.title('Wave Form of ' + filename)
- # pyplot.show()
- mfcc_result = mfcc(y=audio, sr=sample_rate, n_mfcc=num_mfcc)
- mean_mfcc = np.mean(mfcc_result, axis=1)
- print("end extract_mfcc time:" + filename + str(datetime.now()))
- return mean_mfcc
- def compute_similarity(feature_1, feature_2):
- print("start compute_similarity time:" + str(datetime.now()))
- distance = euclidean(feature_1, feature_2)
- result = (100-distance)/100
- # 以下2种弃用的计算方式
- # result = 1-(distance*distance / 10000)
- # result = 1 - np.square((100-distance)/100)
- if result < 0:
- result = 0
- print("end compute_similarity time:" + str(datetime.now()))
- return result
|
