<![CDATA[Speech signals play a crucial role in human communication, particularly in speech recognition systems. However, speech recognition performance is often compromised by noise in the audio signal. This study aims to examine the effect of wavelet denoising technique on greeting sentence data containing artificial white noise before performing speech recognition using Long Short-Term Memory (LSTM). Mel Frequency Cepstral Coefficient (MFCC) is used as speech feature extraction. The results show that speech recognition accuracy reaches 90% on clean data. Accuracy drops to 51% when tested on data with noise, indicating a significant decrease of 39 percentage points. After applying the wavelet denoising method, accuracy improved using the two best parameter combinations. The combination with the highest SNR value resulted in an improvement of 18 percentage points, while the combination with the highest PESQ value resulted in an improvement of 13 percentage points. These findings indicate that the wavelet denoising method is capable of improving the performance of LSTM-based speech recognition in noisy environments.]]>
Copyrights © 2026
