Plotting TSNE scatter plot for speech spectrograms

Question

I have spectrogram of speech waveforms belonging to 4 classes. I want to plot TSNE scatter plot to visualize the distribution of the speech files amongst the four classes. How can I do it with tsne ?

Answer 1

Say you have your spectrograms data as an array of shape (n_points, n_dims) and your associated labels.

Here I will generate mines :

import matplotlib.pyplot as plt
import numpy as np
from sklearn.manifold import TSNE

n_points = 50
n_classes = 4
n_dims = 150

# Generate data
labels = np.random.randint(n_classes, size=n_points)
X = np.random.normal(size=(n_points, n_dims))
X = (X.transpose() + labels).transpose()

Then you can simply apply TSNE on your data to make it bidimensional and plot it.

# Do TSNE
X_embedded = TSNE(n_components=2).fit_transform(X)

# Plot
names = ['class_1', 'class_2', 'class_3', 'class_4']
for i in range(n_classes):
    X_label = X_embedded[np.where(labels == i)]
    plt.scatter(X_label[:, 0], X_label[:, 1], label=names[i])
plt.legend()