Extended Abstract: Neural-Network Based Auditory Feedback Modulation in Real-Time

An impaired voice quality (i.e., hoarseness) is a key symptom of voice disorders, including functional dysphonia. Auditory feedback modulation (AFM) of voice quality, in which patients' existing hoarseness is intensified through headphones, could potentially offer a noninvasive treatment approach by encouraging compensatory vocal adjustments that improve voice quality. This study aims to artificially induce hoarseness using neural networks (NNs) for the modification of the cepstrum. Since cepstral parameters like cepstral peak prominence (CPP), jitter, and shimmer reflect voice quality, we propose that targeted cepstral modifications could be used to simulate hoarseness. A multilayer perceptron (MLP), as a very simple form of a neural network, ensures real-time processing, paving the way for using artificial intelligence in future AFM applications in speech therapy and training.