In-clinic and tele-rehabilitation paths for multiple sclerosis patients with speech disorders - Comparison through vocal signals acquired by in-air and contact microphones.

Voice is essential for communication, relationship building, and emotional expression: its production involves the vocal cords, respiratory system, and upper vocal tract, and improper control can lead to dysphonia. Multiple Sclerosis (MS) is an immune-mediated demyelinating disease of the central nervous system that can result in physical disabilities, including voice disorders, in fact 62% of MS patients experience hypophonia. The Don Gnocchi Foundation supports patient’s rehabilitation, mainly with the Lee Silverman Voice Treatment (LSVT)-Loud, an effective therapy based on motor learning and neural plasticity principles. Traditionally offered in person, LSVT-Loud has recently transitioned online due to COVID-19, allowing patients to receive therapy at home without needing transportation or accompaniment, which improves accessibility and convenience. Therefore, given the advantages of online rehabilitation for patients, it has been decided to focus this study on verifying the non-inferiority of tele-rehabilitation compared to in-person one. To address this objective, the Don Carlo Gnocchi Foundation provided a dataset of 20 MS patients, 10 treated with tele-rehabilitation and 10 with in-clinic rehabilitation. Each patient is asked to produce three repetitions of the sustained vowel /a/, a monologue, and a reading of the poem 'Notturno' at three different times T0 (pre), T1(post), and T2 (follow-up). This vocal material is recorded using two different devices, a vocal recorder equipped with an in-air microphone and the Vocal Holter (VH) device, which include a contact microphone. In addition to the objective of assessing the non-inferiority of tele-rehabilitation, this study also aims to compare the two different recording systems. The recordings of in-air microphone are first cleaned removing artefacts using Audacity, then they are pre-processed in MATLAB and eventually the vocal features are extracted. The main parameters extracted for comparison between VH and in-air microphone, or tele-rehabilitation and in-clinc rehabilitation, include fundamental frequency, jitter, shimmer, cepstral peak prominence smoothed (CPPS), harmonic-to-noise ratio (HNR), and sound pressure level (SPL). In this study, unlike previous work, the pre-processing phase uses varying thresholds to differentiate between harmonic and non-harmonic frames. While a 0 dB threshold is previously used for healthy subjects, thresholds of -10 dB and -20 dB are investigated for MS patients to account for disease-specific non-harmonic frames. To demonstrate the results evaluated with different thresholds, the mean value of HNR at T0 is computed, obtaining for a 0 dB threshold the value of 7.8 dB. Compared to this value, using the -10 dB and -20 dB thresholds results in HNR differences of -1.8 dB and -1.9 dB respectively. In the case of the VH, it generates a text file containing all the parameters extracted during the recording. By analyzing parameters trends across different time it is possible to assess whether rehabilitation has led to improvement and to determine which of the two techniques is more effective or if they have comparable effects. Among the analyzed parameters, CPPS shows the most similar mean values between the presence and tele-rehabilitation conditions: the mean difference between T1 and T0 for rehabilitation in presence is 0.33 dB, while in the online case it is 0.35 dB. As a result, it has been demonstrated that tele-rehabilitation and in-clinc rehabilitation are comparable, showing similar benefits.