在本文中，我将展示如何使用张量流来预测音乐风格。
在我的示例中，我比较了电子音乐和古典音乐。

您可以在我的github上找到代码：
https://github.com/victordalet/sound_to_partition

I - 数据集

第一步，您需要创建一个数据集文件夹，并在里面添加一个音乐风格文件夹，例如我添加一个 techno 文件夹和 classic 文件夹，其中放置我的 wav 歌曲。

II - 火车

我创建一个训练文件，参数 max_epochs 需要完成。

修改构造函数中与数据集文件夹中的目录对应的类。

在加载和处理方法中，我从不同的目录检索wav文件并获取频谱图。

出于训练目的，我使用 Keras 卷积和模型。

import os
import sys
from typing import List

import librosa
import numpy as np
from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, Flatten, Dense
from tensorflow.keras.models import Model
from tensorflow.keras.optimizers import Adam
from sklearn.model_selection import train_test_split
from tensorflow.keras.utils import to_categorical
from tensorflow.image import resize



class Train:

    def __init__(self):
        self.X_train = None
        self.X_test = None
        self.y_train = None
        self.y_test = None
        self.data_dir: str = 'dataset'
        self.classes: List[str] = ['techno','classic']
        self.max_epochs: int = int(sys.argv[1])

    @staticmethod
    def load_and_preprocess_data(data_dir, classes, target_shape=(128, 128)):
        data = []
        labels = []

        for i, class_name in enumerate(classes):
            class_dir = os.path.join(data_dir, class_name)
            for filename in os.listdir(class_dir):
                if filename.endswith('.wav'):
                    file_path = os.path.join(class_dir, filename)
                    audio_data, sample_rate = librosa.load(file_path, sr=None)
                    mel_spectrogram = librosa.feature.melspectrogram(y=audio_data, sr=sample_rate)
                    mel_spectrogram = resize(np.expand_dims(mel_spectrogram, axis=-1), target_shape)
                    data.append(mel_spectrogram)
                    labels.append(i)

        return np.array(data), np.array(labels)

    def create_model(self):
        data, labels = self.load_and_preprocess_data(self.data_dir, self.classes)
        labels = to_categorical(labels, num_classes=len(self.classes))  # Convert labels to one-hot encoding
        self.X_train, self.X_test, self.y_train, self.y_test = train_test_split(data, labels, test_size=0.2,
                                                                                random_state=42)

        input_shape = self.X_train[0].shape
        input_layer = Input(shape=input_shape)
        x = Conv2D(32, (3, 3), activation='relu')(input_layer)
        x = MaxPooling2D((2, 2))(x)
        x = Conv2D(64, (3, 3), activation='relu')(x)
        x = MaxPooling2D((2, 2))(x)
        x = Flatten()(x)
        x = Dense(64, activation='relu')(x)
        output_layer = Dense(len(self.classes), activation='softmax')(x)
        self.model = Model(input_layer, output_layer)

        self.model.compile(optimizer=Adam(learning_rate=0.001), loss='categorical_crossentropy', metrics=['accuracy'])

    def train_model(self):
        self.model.fit(self.X_train, self.y_train, epochs=self.max_epochs, batch_size=32,
                       validation_data=(self.X_test, self.y_test))
        test_accuracy = self.model.evaluate(self.X_test, self.y_test, verbose=0)
        print(test_accuracy[1])

    def save_model(self):
        self.model.save('weight.h5')


if __name__ == '__main__':
    train = Train()
    train.create_model()
    train.train_model()
    train.save_model()

III-测试

为了测试和使用该模型，我创建了此类来检索权重并预测音乐的风格。

不要忘记将正确的类添加到构造函数中。

from typing import List

import librosa
import numpy as np
from tensorflow.keras.models import load_model
from tensorflow.image import resize
import tensorflow as tf



class Test:

    def __init__(self, audio_file_path: str):
        self.model = load_model('weight.h5')
        self.target_shape = (128, 128)
        self.classes: List[str] = ['techno','classic']
        self.audio_file_path: str = audio_file_path

    def test_audio(self, file_path, model):
        audio_data, sample_rate = librosa.load(file_path, sr=None)
        mel_spectrogram = librosa.feature.melspectrogram(y=audio_data, sr=sample_rate)
        mel_spectrogram = resize(np.expand_dims(mel_spectrogram, axis=-1), self.target_shape)
        mel_spectrogram = tf.reshape(mel_spectrogram, (1,)   self.target_shape   (1,))

        predictions = model.predict(mel_spectrogram)

        class_probabilities = predictions[0]

        predicted_class_index = np.argmax(class_probabilities)

        return class_probabilities, predicted_class_index

    def test(self):
        class_probabilities, predicted_class_index = self.test_audio(self.audio_file_path, self.model)

        for i, class_label in enumerate(self.classes):
            probability = class_probabilities[i]
            print(f'Class: {class_label}, Probability: {probability:.4f}')

        predicted_class = self.classes[predicted_class_index]
        accuracy = class_probabilities[predicted_class_index]
        print(f'The audio is classified as: {predicted_class}')
        print(f'Accuracy: {accuracy:.4f}')