Big Voice Data

Voice is an important aspect of human communication. It plays a central role in our social interactions, is a crucial tool in many professions, and is typically used for purposes of artistic expression. From both medical and educational perspectives, it is therefore important to precisely understand the process of voice production. The principles are well-known: Air flows through the trachea and causes the vocal folds (colloquially known as "vocal cords") in the larynx to vibrate at frequencies ranging from 50 to well over 1000 times per second, depending on the type of vocalization. This creates the primary vocal sound, which is further acoustically modified in the throat and oral cavity. There is thus a direct and causal connection between the nature of vocal fold vibration and the quality of the audible voice. This insight is particularly significant for voice therapy: A hoarse or diseased voice is often the result of disturbed vocal fold vibration. Although much knowledge about voice production already exists, there are still significant unanswered questions. Previous research has been limited by several challenges and technical constraints: 1. Often only the average speaking voice range is studied. Other voice ranges, which are important for singing, shouting, and nonverbal vocal communication, are significantly less researched. 2. Because the vocal folds vibrate several hundred times per second, special high-speed cameras are required to adequately document their vibration. However, this generates several thousand images per second, resulting in extremely large amounts of data. Analysis of these data requires powerful methods from the field of artificial intelligence that have only become available in recent years. 3. Due to the high technical effort involved, many studies focus only on a small group of participants. Systematic and comprehensive studies using high-speed video recordings on a larger number of persons are yet outstanding. The current understanding of human voice production is therefore incomplete, which represents a significant limitation for current voice pedagogical and medical concepts. This is precisely where the current project, conducted at the Mozarteum University Salzburg and Trier University of Applied Sciences, comes in to close existing research gaps. Using state-of-the- art high-speed endoscopy, the entire vocal range of a larger group of people is being examined. AI- supported algorithms developed specifically for this project enable a comprehensive analysis of the relationship between vocal fold vibration and voice quality. Upon completion of the project, all data collected will be published as an open-source dataset, allowing the world-wide scientific community to conduct further research.