Tensorflow 音樂預測

在本文中，我將展示如何使用張量流來預測音樂風格。
在我的範例中，我比較了電子音樂和古典音樂。

你可以在我的github上找到程式碼：
https://github.com/victordalet/sound_to_partition

---

I - 資料集

第一步，您需要建立一個資料集資料夾，並在裡面新增一個音樂風格資料夾，例如我新增一個 techno 資料夾和 classic 資料夾，其中放置我的 wav 歌曲。

II - 火車

我建立一個訓練文件，並使用要完成的參數 max_epochs。

修改建構函式中與資料集資料夾中的目錄對應的類別。

在載入和處理方法中，我從不同的目錄檢索wav檔案並取得頻譜圖。

出於訓練目的，我使用 Keras 卷積和模型。

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-測試

為了測試和使用該模型，我創建了此類來檢索權重並預測音樂的風格。

不要忘記將正確的類別加入建構函式中。

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}')

以上是Tensorflow 音樂預測的詳細內容。更多資訊請關注PHP中文網其他相關文章！