The development of virtual assistants, speech processing is largely used in various industrial and domestic applications and services. Pre-processing tasks like noise reduction and silence removal are necessary in order to process speech signals. In order to sustain speech emotions, this research focuses on keeping the inner silence present in speech signals rather than deleting it. Short-Term Energy (STE) and Zero-Crossing Rate (ZCR) are the two common approaches for silence removal. In order to eliminate the embedded silence in an audio file, threshold-based techniques produce promising results. However, such threshold-based approaches do not perform effectively on all types of audio formats. The performance of the suggested approach is confirmed to consistently produce acceptable and reliable results on the given audio/speech signals. MATLAB was chosen as the implementation tool. The metrics used for the evaluation were duration and size of the processed audio files. The authors have proposed novel threshold-based based method that uses Root Mean Square Energy (RMSE) for silence trimming. This article describes the mechanism and merits of the proposed method.