Automatic Speech Recognition (ASR) research has made significant strides with the adoption of deep learning techniques. This paper examines the development and performance of three ASR models: a CTC-based model using 2D Convolutional Neural Networks (CNN) and Recurrent Neural Networks (RNN), a sequence-to-sequence Transformer model, and OpenAI’s Whisper model. These models are evaluated using the LJ Speech Dataset, consisting of 13,100 English audio clips, with corresponding transcriptions. A comparative analysis reveals that the CTC-based model achieves a Word Error Rate (WER) of 0.1429, a Character Error Rate (CER) of 0.0308 and a Real-Time Factor (RTF) of 1. The CTC-based model effectively combines CNNs for feature extraction and RNNs for sequence modeling, using Connectionist Temporal Classification (CTC) loss for alignment and prediction without the need for pre-segmented data. Further evaluation compares the Transformer and Whisper models to determine the most suitable approach for different ASR applications. This study provides an indepth analysis of the trade-offs between these models, offering insights into their strengths, limitations, and suitability for various ASR tasks.