Stuttering is a speech disorder characterized by disruptions in the fluency and flow of speech. This paper introduces Detecting Different types of Stutter (DDoS), a novel deep neural network (DNN) approach for identifying and categorizing various stutter disfluencies. Our study presents a model that relies solely on spectral features as the foundation of our approach. DDoS effectively extracts features like MFCC, LPC and LPCC and their combination and feeds them into a TDNN architecture. The model captures patterns in speech signals while considering shift-invariance and contextual modelling. We evaluated our model on the Uclass and LibriStutter dataset, which include a significant number of females \((\approx 40)\) and males \((\approx 120)\) participants. Unlike other studies focusing on different deep learning models, we prioritize using multiple features and their combinations. Consequently, our model achieves superior results and outperforms the state-of-the-art TDNN based approach in various aspects.