依赖于OpenAi问答得到的程序，需要继续调优

# 用于技术测试，不作为功能代码
import pywt
import librosa
import numpy as np
from scipy.spatial.distance import cosine, correlation

import tensorflow as tf

import librosa
import numpy as np


# 定义一个函数，用于计算两个音频信号之间的相似度分数
def calculate_similarity(audio_file1, audio_file2):
    # 加载音频文件，并提取多种不同的特征
    audio1, sr1 = librosa.load(audio_file1)
    audio2, sr2 = librosa.load(audio_file2)
    chroma1 = librosa.feature.chroma_stft(audio1, sr=sr1)
    chroma2 = librosa.feature.chroma_stft(audio2, sr=sr2)
    mfcc1 = librosa.feature.mfcc(audio1, sr=sr1)
    mfcc2 = librosa.feature.mfcc(audio2, sr=sr2)

    # 计算特征之间的相似度分数
    chroma_similarity = 1 - np.mean(librosa.feature.delta(chroma1) * librosa.feature.delta(chroma2))
    mfcc_similarity = 1 - cosine(np.mean(mfcc1, axis=1), np.mean(mfcc2, axis=1))
    correlation_similarity = 1 - np.mean(correlation(chroma1, chroma2))

    # 计算最终的相似度分数
    similarity = (chroma_similarity + mfcc_similarity + correlation_similarity) / 3.0

    return similarity



# 加载两个音频文件


# 定义一个函数，用于计算两个音频信号之间的相似度分数
def calculate_similarity1(audio_file1, audio_file2):

    # 加载音频文件，并提取多种不同的特征
    audio1, sr1 = librosa.load(audio_file1, sr=16000)
    audio2, sr2 = librosa.load(audio_file2, sr=16000)

    # 如果采样率不同，将其转换为相同的采样率
    if sr1 != sr2:
        audio2 = librosa.resample(audio2, sr2, sr1)
        sr2 = sr1

    # 对音频信号进行短时傅里叶变换（STFT）和滑动窗口分帧处理，获取音频特征图
    stft1 = tf.signal.stft(audio1, frame_length=1024, frame_step=256, pad_end=True)
    stft2 = tf.signal.stft(audio2, frame_length=1024, frame_step=256, pad_end=True)
    spectrogram1 = tf.abs(stft1)
    spectrogram2 = tf.abs(stft2)
    print(spectrogram1.shape)
    spectrogram1 = tf.expand_dims(spectrogram1, axis=-1)
    print(spectrogram1.shape)
    spectrogram2 = tf.expand_dims(spectrogram2, axis=-1)
    print(spectrogram2.shape)

    # 使用卷积神经网络（CNN）对音频特征图进行特征提取
    model = tf.keras.Sequential([
        tf.keras.layers.Input(shape=(spectrogram1.shape[0], spectrogram1.shape[1], spectrogram1.shape[2])),
        tf.keras.layers.Conv2D(16, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.MaxPooling2D((2, 2)),
        tf.keras.layers.Conv2D(32, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.MaxPooling2D((2, 2)),
        tf.keras.layers.Conv2D(64, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(128, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(64, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(32, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(16, activation='relu'),
        tf.keras.layers.Dense(8, activation='relu'),
    ])
    model.build(input_shape=(None, spectrogram1.shape[0], spectrogram1.shape[1], spectrogram1.shape[2]))
    model.summary()
    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.01),
                  loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
                  metrics=['accuracy'])
    # history = model.fit(epochs=5)
    model.save('model.h5')
    # 加载预训练模型参数
    model.load_weights('model.h5')

    # 对音频信号进行预测，获取音频特征向量
    print("变化后的shape{}".format(spectrogram1.shape))
    spectrogram1 = tf.expand_dims(spectrogram1, axis=0)
    spectrogram2 = tf.expand_dims(spectrogram2, axis=0)
    print(spectrogram1.shape)
    features1 = model.predict(spectrogram1)
    features2 = model.predict(spectrogram2)

    # 计算特征向量之间的余弦相似度分数
    similarity = np.dot(features1, features2.T) / (np.linalg.norm(features1) * np.linalg.norm(features2))

    return similarity

# base_41 = 
# base_39 = 
# res = calculate_similarity1(base_41, base_41)
# print(res)


def calculate_similarity2(audio_file1, audio_file2):
    """
    比较音频的相似度
    Args:
        audio_file1: 短的标准=音
        audio_file2: 长的录制音

    Returns:

    """

    # 加载音频文件，并提取多种不同的特征
    audio1, sr1 = librosa.load(audio_file1, sr=16000)
    audio2, sr2 = librosa.load(audio_file2, sr=16000)

    # 如果采样率不同，将其转换为相同的采样率
    if sr1 != sr2:
        audio2 = librosa.resample(audio2, sr2, sr1)
        sr2 = sr1

    # 对音频信号进行短时傅里叶变换（STFT）和滑动窗口分帧处理，获取音频特征图
    stft1 = tf.signal.stft(audio1, frame_length=1024, frame_step=256, pad_end=True)
    stft2 = tf.signal.stft(audio2, frame_length=1024, frame_step=256, pad_end=True)
    spectrogram1 = tf.abs(stft1)
    spectrogram2 = tf.abs(stft2)
    spectrogram1 = tf.expand_dims(spectrogram1, axis=-1)
    spectrogram2 = tf.expand_dims(spectrogram2, axis=-1)
    print("输入的shape:")
    print(spectrogram1.shape)
    print(spectrogram2.shape)

    # 使用卷积神经网络（CNN）对音频特征图进行特征提取
    model = tf.keras.Sequential([
        tf.keras.layers.Input(shape=(spectrogram1.shape[0], spectrogram1.shape[1], spectrogram1.shape[2])),
        tf.keras.layers.Conv2D(16, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.MaxPooling2D((2, 2)),
        tf.keras.layers.Conv2D(32, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.MaxPooling2D((2, 2)),
        tf.keras.layers.Conv2D(64, (3, 3), activation='relu', padding='same'),
        tf.keras.layers.Flatten(),
        tf.keras.layers.Dense(128, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(64, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(32, activation='relu'),
        tf.keras.layers.Dropout(0.5),
        tf.keras.layers.Dense(16, activation='relu'),
        tf.keras.layers.Dense(8, activation='relu'),
    ])
    model.build(input_shape=(None, spectrogram1.shape[0], spectrogram1.shape[1], spectrogram1.shape[2]))
    model.summary()
    model.compile(optimizer=tf.keras.optimizers.Adam(learning_rate=0.01),
                  loss=tf.keras.losses.CategoricalCrossentropy(from_logits=True),
                  metrics=['accuracy'])
    model.save('model.h5')
    # 加载预训练模型参数
    model.load_weights('model.h5')
    # 计算第二个音频的STFT特征的相似度
    similarity_scores = []
    window_size = spectrogram1.shape[0]
    print("nihao")
    print(window_size)
    hop_length = window_size // 2
    spectrogram1 = tf.expand_dims(spectrogram1, axis=0)
    stand_feature1 = model.predict(spectrogram1)
    print(spectrogram1.shape)
    print(spectrogram2.shape)
    for i in range(0, spectrogram2.shape[1] - window_size):
        window = spectrogram2[i:i + window_size, :, :]
        print("Shape========================")
        print(spectrogram1.shape)
        print(window.shape)
        f2 = model.predict(window[np.newaxis, ...])
        similarity = np.dot(stand_feature1, f2.T) / (np.linalg.norm(stand_feature1) * np.linalg.norm(f2))
        print("相似度：{}".format(similarity))
        similarity_scores.append(similarity.tolist())
    return similarity_scores

#  已经快最近最优的解了
res_list = calculate_similarity2(base_,base_41)

print(res_list)