Music Genre Classifier using Machine Learning

Music is the art of arranging sound and noise together to create harmony, melody, rhythm, and expressive content. It is organized so that humans and sometimes other living organisms can express their current emotions with it.

We all have our own playlist, which we listen to while traveling, studying, dancing, etc.

In short, every emotion has a different genre. So here today, we will study how can we implement the task of genre classification using Machine Learning in Python.

Before starting the code, download the data from this link.

Let’s start with the code.

Import Libraries and Dataset

Firstly we need to import Libraries :

• Pandas: To import files/datasets.
• Matplotlib: To visualize the data frame.
• Numpy: To perform operations like scaling and correlation.
• Seaborn: To visualize the data frame.
• Librosa: To visualize the audio data. Install this library by “pip install librosa” command.

Python3

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import librosa.display

Now to import the data file run the below command.

Python3

music_data = pd.read_csv('file.csv')
music_data.head(5)

Output :

Exploratory Data Analysis

Let’s find out the count of each music label.

Python3

music_data['label'].value_counts()

Output:

blues        100
classical    100
country      100
disco        100
hiphop       100
jazz         100
metal        100
pop          100
reggae       100
rock         100

We can also analysis the sound waves of the audio using the Librosa library.

Let’s visualize few of them with the below code.

Python3

path = 'genres_original/blues/blues.00000.wav'
plt.figure(figsize=(14, 5))
x, sr = librosa.load(path)
librosa.display.waveplot(x, sr=sr)
id.Audio(path)

print("Blue")

Output : 

Blue

Python3

path = 'genres_original/metal/metal.00000.wav'
plt.figure(figsize=(14, 5))
x, sr = librosa.load(path)
librosa.display.waveplot(x, sr=sr,color='orange')
id.Audio(path)

print("Metal")

Output : 

Metal

Python3

path = 'genres_original/pop/pop.00000.wav'
plt.figure(figsize=(14, 5))
x, sr = librosa.load(path)
librosa.display.waveplot(x, sr=sr,color='purple')
id.Audio(path)

print("Pop")

Output : 

Pop

Python3

path = 'genres_original/hiphop/hiphop.00000.wav'
plt.figure(figsize=(14, 5))
x, sr = librosa.load(path)
librosa.display.waveplot(x, sr=sr,color='grey')
id.Audio(path)

print("HipHop")

Output : 

HipHop

Python3

import numpy as np
import seaborn as sns

# Computing the Correlation Matrix
spike_cols = [col for col in data.columns if 'mean' in col]

# Set up the matplotlib figure
f, ax = plt.subplots(figsize=(16, 11));

# Draw the heatmap with the mask and correct aspect ratio
sns.heatmap(data[spike_cols].corr(), cmap='YlGn')

plt.title('Heatmap for MEAN variables', fontsize = 20)
plt.xticks(fontsize = 10)
plt.yticks(fontsize = 10);

Output : 

Data Preprocessing 

Initially, we need to use LabelEncoder() to convert the labels into integer.

Python3

from sklearn import preprocessing
label_encoder = preprocessing.LabelEncoder()
music_data['label'] = label_encoder.fit_transform(music_data['label'])

As filename column is not a relevant, so we can drop it.

Python3

X = music_data.drop(['label','filename'],axis=1)
y = music_data['label']

Now the data needs to be scaled, to make the model more stable and train fast.

Python3

cols = X.columns
minmax = preprocessing.MinMaxScaler()
np_scaled = minmax.fit_transform(X)

# new data frame with the new scaled data.
X = pd.DataFrame(np_scaled, columns = cols)

Model Training 

Initially, split the model using train_test_split module. 

Python3

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                    test_size=0.3,
                                                    random_state=111)
X_train.shape, X_test.shape, y_train.shape, y_test.shape

We will be testing our datasets on below models : 

• K-Neighbors Classifier :  KNeighborsClassifier looks for topmost n_neighbors using different distance methods like Euclidean distance.
• Decision Tree Classifier : In Decision tree each node is trained by splitting the data is continuously according to a certain parameter.
• Random Forest : Random Forest Classifier fits a number of decision tree classifiers on many sub-samples of the dataset and then use the average to improve the results.
• Logistics Regression : Logistic Regression is a regression model that predicts the probability of a given data belongs to the particular category or not.
• Cat Boost : CatBoost implements decision trees and restricts the features split per level to one, which help in decreasing prediction time. It also handles categorical features effectively.
• Gradient Boost : In Gradient Boost an decision trees are implemented in a sequential manner which enhance the performance.

Python3

from sklearn.metrics import accuracy_score
from sklearn.neighbors import KNeighborsClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.linear_model import LogisticRegression
import catboost as cb
from xgboost import XGBClassifier

rf = RandomForestClassifier(n_estimators=1000, max_depth=10, random_state=0)
cbc = cb.CatBoostClassifier(verbose=0, eval_metric='Accuracy', loss_function='MultiClass')
xgb = XGBClassifier(n_estimators=1000, learning_rate=0.05)

for clf in (rf, cbc, xgb):
    clf.fit(X_train, y_train)
    preds = clf.predict(X_test)
    print(clf.__class__.__name__,accuracy_score(y_test, preds))

Output :

RandomForestClassifier 0.78
CatBoostClassifier 0.8333333333333334
XGBClassifier 0.7933333333333333

Neural Network

Let’s evaluate the dataset with the simple Neural network.

Python3

import tensorflow.keras as keras
from tensorflow.keras import Sequential
from tensorflow.keras.layers import *

model = Sequential()

model.add(Flatten(input_shape=(58,)))
model.add(Dense(256, activation='relu'))
model.add(BatchNormalization())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.3))
model.add(Dense(10, activation='softmax'))
model.summary()

Output :

Compiling and fitting the model 

Python3

# compile the model
adam = keras.optimizers.Adam(lr=1e-4)
model.compile(optimizer=adam,
             loss="sparse_categorical_crossentropy",
             metrics=["accuracy"])

hist = model.fit(X_train, y_train,
                 validation_data = (X_test,y_test),
                 epochs = 100,
                 batch_size = 32)

100 epochs will take some time.

Once done, then we can do evaluation.

Evaluation

Let’s check the test accuracy by below code.

Python3

test_error, test_accuracy = model.evaluate(X_test, y_test, verbose=1)
print(f"Test accuracy: {test_accuracy}")

Output : 

Test accuracy: 0.7566666603088379

Now we can evaluate the accuracy using line-plots.

Python3

fig, axs = plt.subplots(2,figsize=(10,10))

# accuracy
axs[0].plot(hist.history["accuracy"], label="train")
axs[0].plot(hist.history["val_accuracy"], label="test")
axs[0].set_ylabel("Accuracy")
axs[0].legend()
axs[0].set_title("Accuracy")

# Error
axs[1].plot(hist.history["loss"], label="train")
axs[1].plot(hist.history["val_loss"], label="test")
axs[1].set_ylabel("Error")
axs[1].legend()
axs[1].set_title("Error")

plt.show()

Output :

Conclusion 

Ensemble Learning and Neural nets has been proven the best way for classification of the genre with the accuracy of more than 80%