PuppyServer/aiDogProject_docker/aiDogProject/aiDogApp/wave_compare.py

from datetime import datetime
import os
os.environ["NUMBA_DISABLE_FUNCTION_CACHING"] = "1"
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()
    # 使用librosa.feature.mfcc而不是直接导入mfcc
    mfcc_result = librosa.feature.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()))
    time_start = 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

    time_end = datetime.now()
    time_diff = (time_end - time_start).total_seconds()
    print("end compute_similarity time:" + str(datetime.now()))


    res={
        "result":result,
        "time_diff":time_diff
    }
    return res