Determinación Dinámica de Segmentos Vocálicos Basada en Métodos Estadísticos

PAOLA BUSTAMANTE; NATALIA LÓPEZ; ELISA PEREZ; LUCÍA QUINTERO

BUCARAMANGA

Congreso; VII Congreso Latinoamericano de Ingeniería Biomédica ? CLAIB; 2016

Universidad Autónoma de Bucaramanga

The detection of voice activity in the area of voice recognition had a great development in different applications, for example in noise reduction devices in digital hearing aids or in the eliminating unnecessary transmission of segments no voice in encoding voice. The methods used to achieve good detection depend on the scope and robustness to noise to be reached. In this paper, we used a statistical method based on the criterion of Neyman Pearson to develop a segmentation algorithm that defines a dynamic threshold. This binary threshold is obtained through the statistical parameters of the non-speech segments and also it allows differentiate between speech and non-speech segments. Then an error analysis was performed by calculating the signal to noise ratio in order to evaluate the performance of the detector, then it was com-pared with a commercial detector. The segmentation between voice and non-voice segments allows to improve the quality of the features that are extracted from the speech segments and remove noise that comes with silence.