Early diagnosis of dementia is crucial for mitigating the consequences of this disease in patients. Previous studies have demonstrated that it is possible to detect the symptoms of dementia, in some cases even years before the onset of the disease, by detecting neurodegeneration-associated characteristics in a person's speech. This paper presents an automatic method for detecting dementia caused by Alzheimer's disease (AD) through a wide range of acoustic and linguistic features extracted from the person's speech. Two well-known databases containing speech for patients with AD and healthy controls are used to this end: DementiaBank and ADReSS. The experimental results show that our system is able to achieve state-of-theart performance on both databases. Furthermore, our results also show that the linguistic features extracted from the speech transcription are significantly better for detecting dementia.

No abstract

ReSSInt aims at investigating the use of silent speech interfaces (SSIs) for restoring communication to individuals who have been deprived of the ability to speak. SSIs are devices which capture non-acoustic biosignals generated during the speech production process and use them to predict the intended message. Two are the biosignals that will be investigated in this project: electromyography (EMG) signals representing electrical activity driving the facial muscles and invasive electroencephalography (iEEG) neural signals captured by means of invasive electrodes implanted on the brain. From the whole spectrum of speech disorders which may affect a person's voice, ReSSInt will address two particular conditions: (i) voice loss after total laryngectomy and (ii) neurodegenerative diseases and other traumatic injuries which may leave an individual paralyzed and, eventually, unable to speak. To make this technology truly beneficial for these persons, this project aims at generating intelligible speech of reasonable quality. This will be tackled by recording large databases and the use of state-of-the-art generative deep learning techniques. Finally, different voice rehabilitation scenarios are foreseen within the project, which will lead to innovative research solutions for SSIs and a real impact on society by improving the life of people with speech impediments.

This paper presents a decision-making algorithm based on a modified version of the Bellman Equation to deal with Electromagnetic Interference errors in a communication channel inside a harsh electromagnetic reverberant environment.The Bellman equation is a fundamental concept in decisionmaking problems such as Markov Decision Processes. Such processes model decision-making in situations where the outcomes are partly random and partly controlled by an agent (decisionmaker).Recent studies have implemented Markov Decision Processes as a tool for risk assessment in areas such as robotics or aviation. However, so far, no research has been reported that uses the Bellman Equation or Markov Decision Processes to deal with risks related to electromagnetic disturbances.In our study, a wired communication channel that uses Non-Return-to-Zero-Level data encoding and Hamming code for error detection and correction is disturbed. First, the packet error rate is calculated and compared with and without the proposed algorithm for different electromagnetic disturbance frequencies and bit rates. The gain is compared at different packet error rates when the algorithm's parameters (called rewards) are optimized and when these are set as random. Last, the influence of the rewards and the maximum number of resends on the algorithm's performance is also studied.

Articulatory-to-acoustic (A2A) synthesis refers to the generation of audible speech from captured movement of the speech articulators. This technique has numerous applications, such as restoring oral communication to people who cannot longer speak due to illness or injury. Most successful techniques so far adopt a supervised learning framework, in which timesynchronous articulatory-and-speech recordings are used to train a supervised machine learning algorithm that can be used later to map articulator movements to speech. This, however, prevents the application of A2A techniques in cases where parallel data is unavailable, e.g., a person has already lost her/his voice and only articulatory data can be captured. In this work, we propose a solution to this problem based on the theory of multi-view learning. The proposed algorithm attempts to find an optimal temporal alignment between pairs of nonaligned articulatory-and-acoustic sequences with the same phonetic content by projecting them into a common latent space where both views are maximally correlated and then applying dynamic time warping. Several variants of this idea are discussed and explored. We show that the quality of speech generated in the non-aligned scenario is comparable to that obtained in the parallel scenario.