Bearing fault diagnosis is a fundamental component of machinery health monitoring, which is used to detect operational abnormalities or faults in bearings. Furthermore, bearing is an essential element in most mechanical systems. This paper proposes bearing fault classification using a novel framework that extracts Short-Time Fourier Transform (STFT) based features using efficient parallel convolution blocks. As such, the proposed model exhibits improved feature extraction, along with multiscale characteristics. Moreover, the model converges rapidly, and within 10 epochs, achieving 99.96% validation accuracy on the University of Ottawa (UOO) dataset. Results are also evaluated using the confusion matrix and ROC in terms of class-wise accuracies.