Resumo

RAMIREZ SANCHEZ, José Manuel; MONTALVO BEREAU, Ana Rosa  e  CALVO DE LARA, José Ramón. Evaluation of Acoustic Features for the Automatic Speech Recognition in Noise Scenarios using Kaldi. EAC [online]. 2019, vol.40, n.3, pp. 51-71.  Epub 08-Set-2019. ISSN 1815-5928.

The present investigation will evaluate the impact of Mel Frequency Cepstral Coefficients (MFCC) and the Perceptual Linear Predictors (PLP) coefficients, in the word error rate (WER) of systems dedicated to Automatic Speech Recognition (ASR). The experimentation will be done with voice signals in Spanish language, in scenarios with unknown noise levels and using the Kaldi state of the art tool. The article concludes by providing evidence in favor of the MFCC as acoustic feature more robust in the task of ASR in noisy scenarios with respect to the PLP; also both features behave similarly in low noise scenarios and the impact of PLP in reducing the time spent by systems dedicated to ASR.

Palavras-chave : Automatic Speech Recognition; Acoustic Features; Kaldi.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )