Detección de actividad de voz en ambiente no ruidoso

Abstract

The current work investigates the detection of voice activity in acoustic signals of French human speech that occur in noise-free environments. It proposes a method capable of differentiating speech and pause events, considering that a pause event that occurs due to the pronunciation of a voiceless stop consonant must be taken as a speech event. For this, it uses components specialized in detection of silence, inhalation and voiceless stop consonants based on Support Vector Machine classifiers (SVM), as well as a component for decision making based on a Finite State Machine. In this sense, the current work shows the drastic impact that the detection of voiceless stop consonants has on the final prediction of speech and pause events.