This study analyses muscular activity during speech production, by modeling the root-mean-square (RMS) levels of electromyographic (EMG) signals. The data belong to a database created to develop an EMG-based silent speech interface (SSI) for alaryngeal speakers. A generalized additive model (GAM) is used, which models non-linear relationships between variables. The results show that EMG signals of silent speech have significantly higher RMS levels than EMG signals of audible speech, suggesting that the speaker compensates for the lack of auditory feedback by articulating more. However, a subsequent qualitative comparison with the patterns associated with alaryngeal speech suggests that the audible speech of laryngeal speakers may be more suitable for developing an SSI for alaryngeal speakers. Further analysis into the different muscles and phonetic outputs indicates that a GAM analysis can be useful in understanding the relationship between muscle use and speech production.