The world is witnessing exponential growth in the field of research in human voice. In the last two decades, laryngology has grown as a discipline in itself becoming independent from otolaryngology. The  evolution of this subspecialty highlights the need for patient-centered evaluation tools, to delve into basic science research, and to have evidence-based voice therapy techniques.

Acoustic and aerodynamic measurements, signal processing, the use of mathematical models, high-speed imaging and artificial intelligence, are all methods that are currently applied in different areas and by different specialties, with
the aim to achieve a better understanding of vocal function.

In the field of Speech-Language Pathology specialized in voice disorders, the study of voice signals is paramount for adequate treatment and follow-up.

This talk proposes to discuss some of the challenges faced in the clinical setting in this new emerging field. We will discuss certain limitations in some of the evaluation tools that are of standard use in a voice center, with the hopes to inspire researchers to continue exploring the clinical usefulness in voice signal processing in the care of the dysphonic patient and bridge the gap between basic science and clinical application.

Physiological modeling has proved a valuable tool for gaining insights into the mechanisms underlying both normal and impaired human phonation. Nowadays, diverse computational models are available to describe and simulate the phenomena and the physical interactions involved in voice production. Two alternatives are especially appealing, combining physiological relevance and low computational burden; the lumped-elements biomechanical modeling of vocal function and the digital waveguide acoustic-tube simulation of the vocal tract. They provide the basis for developing physiologically-based voice synthesizers, thus allowing the investigation of the acoustic effects due to alterations in the glottal function. In recent years, physiological modeling has also drawn the attention of researchers for formulating and solving different inverse problems in the context of phonation. Innovative ideas have arisen from computational methods and models of vocal behavior for estimating relevant clinical information describing the pathophysiology of vocal organs from measured clinical data. The works in our research group focus on combining physiological modeling and signal processing techniques for the clinical assessment of human phonation. In this talk, we will introduce different ideas explored in our group to improve the clinical relevance of low-dimensional models and computational methods well-established for voice assessment. We will discuss a state-space structural representation and analysis scheme for the perturbed pitch period contours in voice signals, new model-based techniques for glottal inverse filtering and acoustic source estimation, a Bayesian estimator of vocal function measures based on the Extended Kalman filter and a biomechanical body-cover model of the vocal folds, and a muscle-controlled physiological voice synthesizer considering the effect of all the five intrinsic laryngeal muscles.

Speech is one of the most natural forms of communication between human beings. Speech signals convey much information about the speaker beyond words, such as emotional and health status, identity, age, gender and even height, to name a few. Human speech communication involves the interaction and coordination of various body parts that produce many signals that can be properly measured and analyzed. Examples of these signals related to the speech production or perception process are muscle signals (EMG), brain activities (EEG, ECoG, fNIRS, MEG, etc.), electroglottogram (EGG) and video recording. The complex processing that takes place at the brain level plays a crucial role in both the production and perception of speech signals. If the cognitive processes involved are better understood, it will be possible to develop more effective artificial methods of communication. Brain signals are also particularly interesting because they are also used for brain-computer interfaces (BCI). In this context, new BCI paradigms related to speech communication were proposed, such as imagined or inner speech. For example, during imagined speech, subjects have to imagine themselves uttering words but without moving muscles nor producing sounds. Some research has been done on the classification of vowels, syllables and whole words using the brain signals acquired during these speech-related paradigms, with promising results for this task that can be called "Speak what you thought". Another interesting line of research achieved is the reconstruction of recognizable speech from the direct recording of listeners' brain signals, thus demonstrating the feasibility of developing systems that can "Hear what you hear". The significant advances of the last decade in computational intelligence, such as the development of several new deep learning architectures together with the availability of the vast amount of experimental data, are responsible for reaching new levels of performance on such challenging tasks. This talk proposes a comprehensive and up-to-date overview of this new emerging field and discusses some of the challenges that need to be addressed to further improve the performance of this technology.