Vocal analysis in multiple sclerosis patients treated with speech therapies - A comparison between in-clinic and remote rehabilitation through vocal indexes provided by different software.

Multiple sclerosis (MS) is a chronic neurological disease that affects the central nervous system, leading to a wide variety of motor, sensory and cognitive impairments. Among these symptoms, speech and voice disorders, such as reduced voice intensity or hypophonia, can significantly affect patients’ communication and overall quality of life. Speech therapy can help reduce the grade of disability: one of the most effective methods is the Lee Silverman Voice Treatment (LSVT-Loud), a therapy focused on improving voice intensity. However, it requires frequent clinic visits, which may limit accessibility for some patients. This study conducts an acoustic analysis aiming to evaluate the effectiveness of LSVT-Loud but also showing the potential benefits of tele-rehabilitation (tele-LSVT-Loud) as an alternative approach. Additionally, it compares results obtained from different software commonly used in vocal analysis (Praat, VOXPlot, MATLAB), trying to assess any discrepancies in the parameters extraction. The study was performed on vocal data provided by the MS Rehabilitation Don Carlo Gnocchi Foundation in Milan. The analysis has been conducted on speech material of 20 MS patients, divided in two groups: one group of 10 patients was treated with standard in-clinic LSVT-Loud therapy, while the other 10 MS patients were selected for the Tele LSVT-Loud program, by accessing a telerehabilitation platform from home. Vocal signals were simultaneously acquired using a vocal recorder equipped with an in-air microphone and the Vocal Holter device that embeds a contact microphone. The dataset includes a set of voice recordings provided for each patient before (T0), after the therapy (T1) and during the follow up (T2). Each set includes three repetitions of the sustained vowel /a/, one minute of free speech and a reading of “Notturno”, an Italian phonetically balanced text. Two indexes were studied to evaluate the effectiveness of both treatments from T0 to T1: the Acoustic Voice Quality Index (AVQI) and the Warning Score (WS). The AVQI ranges from 0 to 10, with lower values indicating healthier voices. This index includes parameters like jitter, shimmer, Cepstral Peak Prominence Smoothed (CPPS), Harmonic-to-Noise Ratio (HNR), Spectral Slope and Tilt, all extracted by a concatenation of 3 seconds of sustained vowel /a/ and a segment of reading. Using the Praat software, the mean AVQI showed a general improvement across all patients; moreover, the comparison between different software highlighted the similarity between Praat and VOXPlot, indeed the mean AVQI calculated with both the software differ of 0.14, resulted negligible. A specific MATLAB script was implemented to evaluate the index, focusing on spectral tilt and slope, which differ from Praat respectively of 3.37% and 0.09%. The WS index, calculated on the average of three repetitions of the sustained vowel, also confirmed the effectiveness of both LSVT therapies, as it generally decreased from T0 to T1, indicating an improvement in vocal health. Further acoustic parameters such as HNR, CPPS, f0, jitter, shimmer and Sound Pressure Level (SPL) were analyzed to assess the treatment efficacy, with comparable outcomes between in-clinic and tele-rehabilitation. Finally, parameters extracted from the Vocal Holter were compared to the other software to validate its performance and identify aspects that need to be improved in future configurations.