cnn_genre_classifier.py

import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import os, json, math, librosa
import IPython.display as ipd
import librosa.display
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Conv2D
import sklearn.model_selection as sk
from sklearn.model_selection import train_test_split

#Klasör adından tür alma.
MUSIC = '/kaggle/input/deneme/Veri/Tür'
music_dataset = [] # Her vaw dosyası için dosya konumları
genre_target = [] 
for root, dirs, files in os.walk(MUSIC):
    for name in files:
        filename = os.path.join(root, name)
        if filename != '/Veri/Tür/Arabesk/Arabesk1.wav':
            music_dataset.append(filename)
            genre_target.append(filename.split("/")[6])
print(set(genre_target))

# Ses Dosyalarını Test Etme
audio_path = music_dataset[150]
x , sr = librosa.load(audio_path)
librosa.load(audio_path, sr=None)
ipd.Audio(audio_path)

# Ses dosyasını bir dalga biçimi olarak görselleştirme
plt.figure(figsize=(16, 5))
librosa.display.waveshow(x, sr=sr)

# Ses dosyasını spektogram olarak görselleştirme
X = librosa.stft(x)
Xdb = librosa.amplitude_to_db(abs(X))
plt.figure(figsize=(14, 5))
librosa.display.specshow(Xdb, sr=sr, x_axis='time', y_axis='hz')
plt.title('Spectogram')
plt.colorbar()

# Sesi Mel-Spectogram Olarak Görselleştirme
file_location = audio_path
y, sr = librosa.load(file_location)
melSpec = librosa.feature.melspectrogram(y=y, sr=sr, n_mels=128)
melSpec_dB = librosa.power_to_db(melSpec, ref=np.max)
plt.figure(figsize=(10, 5))
librosa.display.specshow(melSpec_dB, x_axis='time', y_axis='mel', sr=sr, fmax=8000)
plt.colorbar(format='%+1.0f dB')
plt.title("MelSpectrogram")
plt.tight_layout()
plt.show()

DATASET_PATH = '/kaggle/input/deneme/Veri/Tür'
JSON_PATH = "data_10.json"
SAMPLE_RATE = 22050
TRACK_DURATION = 30 # Saniye cinsinden ölçülür.
SAMPLES_PER_TRACK = SAMPLE_RATE * TRACK_DURATION

def save_mfcc(dataset_path, json_path, num_mfcc=13, n_fft=2048, hop_length=512, num_segments=5):
    """Extracts MFCCs from music dataset and saves them into a json file along witgh genre labels.
        :param dataset_path (str): Path to dataset
        :param json_path (str): Path to json file used to save MFCCs
        :param num_mfcc (int): Number of coefficients to extract
        :param n_fft (int): Interval we consider to apply FFT. Measured in # of samples
        :param hop_length (int): Sliding window for FFT. Measured in # of samples
        :param: num_segments (int): Number of segments we want to divide sample tracks into
        :return:
        """

    # Store mapping(eşleşmeler), labels(etiketler) ve 
    #MFCC ler için bir sözlük oluşturulur.
    data = {
        "mapping": [],
        "labels": [],
        "mfcc": []
    }

    samples_per_segment = int(SAMPLES_PER_TRACK / num_segments)
    num_mfcc_vectors_per_segment = math.ceil(samples_per_segment / hop_length)

    for i, (dirpath, dirnames, filenames) in enumerate(os.walk(dataset_path)):

        if dirpath is not dataset_path:

            semantic_label = dirpath.split("/")[-1]
            data["mapping"].append(semantic_label)
            print("\nProcessing: {}".format(semantic_label))

            for f in filenames:

		# Ses dosyaları yüklenir.

                file_path = os.path.join(dirpath, f)
            
                if file_path != '/kaggle/input/deneme/Veri/Tür/Arabesk/Arabesk1.wav':

                    signal, sample_rate = librosa.load(file_path, sr=SAMPLE_RATE)
            
                    for d in range(num_segments):

                        start = samples_per_segment * d
                        finish = start + samples_per_segment

                        mfcc = librosa.feature.mfcc(signal[start:finish], sample_rate, n_mfcc=num_mfcc, n_fft=n_fft, hop_length=hop_length)
                        mfcc = mfcc.T

                        if len(mfcc) == num_mfcc_vectors_per_segment:
                            data["mfcc"].append(mfcc.tolist())
                            data["labels"].append(i-1)
                            print("{}, segment:{}".format(file_path, d+1))

    with open(json_path, "w") as fp:
        json.dump(data, fp, indent=4)

# Veri Ön İşleme     
save_mfcc(DATASET_PATH, JSON_PATH, num_segments=6)

DATA_PATH = "./data_10.json"


def load_data(data_path):
    """Loads training dataset from json file.
        :param data_path (str): Path to json file containing data
        :return X (ndarray): Inputs
        :return y (ndarray): Targets
    """

    with open(data_path, "r") as fp:
        data = json.load(fp)

    X = np.array(data["mfcc"])
    y = np.array(data["labels"])
    z = np.array(data['mapping'])
    return X, y, z


def plot_history(history):
    """Plots accuracy/loss for training/validation set as a function of the epochs
        :param history: Training history of model
        :return:
    """

    fig, axs = plt.subplots(2)

    # Doğruluk Eğrisi 
    axs[0].plot(history.history["accuracy"], label="train accuracy")
    axs[0].plot(history.history["val_accuracy"], label="test accuracy")
    axs[0].set_ylabel("Accuracy")
    axs[0].legend(loc="lower right")
    axs[0].set_title("Accuracy eval")

    # Hata Eğrisi
    axs[1].plot(history.history["loss"], label="train error")
    axs[1].plot(history.history["val_loss"], label="test error")
    axs[1].set_ylabel("Error")
    axs[1].set_xlabel("Epoch")
    axs[1].legend(loc="upper right")
    axs[1].set_title("Error eval")

    plt.show()


def prepare_datasets(test_size, validation_size):
    """Loads data and splits it into train, validation and test sets.
    :param test_size (float): Value in [0, 1] indicating percentage of data set to allocate to test split
    :param validation_size (float): Value in [0, 1] indicating percentage of train set to allocate to validation split
    :return X_train (ndarray): Input training set
    :return X_validation (ndarray): Input validation set
    :return X_test (ndarray): Input test set
    :return y_train (ndarray): Target training set
    :return y_validation (ndarray): Target validation set
    :return y_test (ndarray): Target test set
    :return z : Mappings for data
    """

    X, y, z = load_data(DATA_PATH)

    X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=test_size)
    X_train, X_validation, y_train, y_validation = train_test_split(X_train, y_train, test_size=validation_size)

    X_train = X_train[..., np.newaxis]
    X_validation = X_validation[..., np.newaxis]
    X_test = X_test[..., np.newaxis]

    return X_train, X_validation, X_test, y_train, y_validation, y_test, z


def build_model(input_shape):
    """Generates CNN model
    :param input_shape (tuple): Shape of input set
    :return model: CNN model
    """

    model = keras.Sequential()

    # 1. Katman
    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu', input_shape=input_shape))
    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))
    model.add(keras.layers.BatchNormalization())

    # 2. Katman
    model.add(keras.layers.Conv2D(32, (3, 3), activation='relu'))
    model.add(keras.layers.MaxPooling2D((3, 3), strides=(2, 2), padding='same'))
    model.add(keras.layers.BatchNormalization())

    # 3. Katman
    model.add(keras.layers.Conv2D(32, (2, 2), activation='relu'))
    model.add(keras.layers.MaxPooling2D((2, 2), strides=(2, 2), padding='same'))
    model.add(keras.layers.BatchNormalization())

    # Özniteliklerin tek boyuta indirgenmesi için kullanılan düzleştirme katmanı
    model.add(keras.layers.Flatten())
    model.add(keras.layers.Dense(64, activation='relu'))
    model.add(keras.layers.Dropout(0.3))

    # Çıktı Katmanı
    model.add(keras.layers.Dense(10, activation='softmax'))

    return model


def predict(model, X, y):
    """Predict a single sample using the trained model
    :param model: Trained classifier
    :param X: Input data
    :param y (int): Target
    """

    X = X[np.newaxis, ...] # array shape (1, 130, 13, 1)

    prediction = model.predict(X)

    predicted_index = np.argmax(prediction, axis=1)
    
    target = z[y]
    predicted = z[predicted_index]

    print("Target: {}, Predicted label: {}".format(target, predicted))

X_train, X_validation, X_test, y_train, y_validation, y_test, z = prepare_datasets(0.25, 0.2)

input_shape = (X_train.shape[1], X_train.shape[2], 1)
model = build_model(input_shape)

optimiser = keras.optimizers.Adam(learning_rate=0.0001)
model.compile(optimizer=optimiser,
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

model.summary()

history = model.fit(X_train, y_train, validation_data=(X_validation, y_validation), batch_size=32, epochs=300)

plot_history(history)

test_loss, test_acc = model.evaluate(X_test, y_test, verbose=2)
print('\nTest accuracy:', test_acc)

#Ses Dosyalarını Test Etme
# Veri setinden tahmin etmek için bir örnek seçilir.
X_to_predict = X_test[1]
y_to_predict = y_test[1]
predict(model, X_to_predict, y_to_predict)