Speech-to-text conversion is facilitated by automatic speech recognition (ASR) systems, which rely on software applications. Traditional ASR methods often involve several distinct components, such as language, acoustic, and pronunciation models, as well as dictionaries. This fragmented approach can be labor-intensive and impact overall performance. This research introduces an enhanced approach to speech recognition by employing an integrated recurrent neural network (RNN) model. The model integrates convolutional neural networks (CNN) with recurrent neural networks (RNN) and employs a connectionist temporal classification (CTC) loss function to optimize performance. Experiments were performed using a dataset of 576,656 sentences (equating to 1240 hours), incorporating erosion techniques for analysis. The model demonstrated strong performance, with a word error rate (WER) of 2%. This method significantly improves ASR systems by eliminating the need for extensive dictionary creation, thereby enhancing both efficiency and accuracy. Future enhancements could include the integration of dialectal and spontaneous speech data to increase the model’s adaptability. Furthermore, tailoring the model to specific tasks could further optimize its performance for targeted applications, thereby enhancing its overall effectiveness.