Noisy environments in various domains pose substantial communication barriers that can negatively impact work efficiency and increase safety risks. With the advancement in robotics technology, there's an increasing need for reliable and efficient voice recognition systems to ensure accuracy and safety during complex operations. This paper introduces a groundbreaking voice recognition method utilizing our innovative smart wearable acoustic sensor based on the ConformerLSTM architecture. The sensing device employs a deep learning model that integrates multiple acoustic features, specifically tailored for processing speech signals captured by acoustic sensors. Our approach involves training the model in both quiet and noisy environmental conditions to enable it to adapt to different environments and achieve approximately 80% recognition accuracy in high-noise settings, significantly outperforming traditional models and enhancing robustness. This study not only advances technology for processing acoustic sensor speech signals but also offers an efficient and reliable solution for speech recognition and synthesis across related fields.