Overview of automatic voice onset measurements

Background: The onset of phonation has been investigated as a biomarker for quantitative voice assessment and identification of phonation type, e.g., breathy, modal, and creaky. Different methods exist for the computation of the voice onset. Furthermore, the onset of phonation can be calculated considering different modalities: acoustic, electroglottographic (EGG), and airflow signals. We investigated existing automatic methods used to measure the voice onset and implemented them into a software suitable for quantitative analysis of recording containing the sustained phonation of vowels.Material and methods: We investigated and implemented three state-of-the-art methods used to measure voice onset: The vocal attack time (VAT), defined as the time lag between the increase of oscillation amplitude of acoustic and EGG signals at voice initiation; the vocal rise time (VRT), defined as the time needed for the amplitude envelope of the acoustic signal to go from 20% to 80% of the maximum value; the voice onset coordination (VOC), defined as the time lag between the increase of oscillation amplitude of EGG and airflow signals at voice initiation. The voice onset measurements were implemented in Python. In addition, a graphical user interface was implemented to measure voice onset.Results: Studies on the voice onset measurements suggest the potential for objective and quantitative clinical assessment. Automatic voice onset measurement can be performed considering different signal processing techniques; however, the type of measure (VAT, VRT, or VOC) depends on the available modality, i.e., acoustic, EGG, or airflow.Discussion: Voice onset measurements provide an objective analysis of abnormal behavior at the start of the phonation process. The main disadvantage of the VAT and VOC measurements is that these can only be computed when two modalities are available, i.e., acoustic and EGG signals for VAT or EGG and airflow signals for VOC. The VRT has the advantage that only the acoustic signal is necessary for computation; however, it depends on arbitrary measurement points to compute the onset.Conclusion: Analysis of vowel production is usually focused on sustained oscillations of the vocal folds, and the transition from the rest state to a saturation value (voice onset) is often overlooked. However, studies in the literature have shown the potential applications of using voice onset measures for clinical evaluation; thus, irregularities in the vocal folds dynamics are yet to be explored