Análise de parâmetros prosódicos em diferentes relações sinal-ruído

Resumo

Speaker identification using speech sample is a common task in forensic phonetics. The presence of noise in the speech sample is a challenge for phoneticians because it can interfere or modify the parameters that are used to identificate the speaker. This study aims to analyze the impact of noise in spontaneous speech by measuring eight prosodic parameters in differents noise-to-ratio signals: high signal-to-noise ratio (recording made in a quiet room) and low signal-to-noise ratio (original recording with adition of gaussian noise). Spectral emphasis and median of fundamental frequency showed differences in its values when the signal-to-noise ratio had changed. The analysis of different parameters related to rhytmical structure of the speech samples were more resistant to noise presence and that parameters seem to be sufficiently robust to be used when the signal-to-noise ratio is low.