Evaluation and comparison using activity signals of speech methods in River Plate spanish using BEPPA Corpus

Abstract

The results obtained after comparing several algorithms which use basic methods of signal processing for speech activity detection of voice or VAD (Voice Activity Detection-VAD), were assessed in order to determine their effectiveness. The algorithms presented in this article are short-time or spectral energy based endpoint detection algorithm, the zero crossing rate method, and the higher order differential (High Order Difference, HOD) method. First, an introduction of the concept of VAD is presented and the need to apply such language algorithms in River Plate is Spanish. Then a summary of the state of the art techniques and algorithms for detecting voice activity is shown with evidence and experiments used to implement algorithms with BEPPA corpus (Evaluation Battery for Patients with Auditive Prostheses, BEPPA – in Spanish).