Toward an automatized classification of /s/-allophones in Guayaquil Spanish

Abstrakt

The paper aims to generate a way of automatized classification of /s/-allophones by finding a correspondence between their acoustic parameters and the qualitative type of the realization. The dataset used for the study comes from CHARG (Corpus de Habla Radiofónica de Guayaquil). The parameters considered for the acoustic description of /s/ are duration (DUR), center of gravity (COG) and percentage of voicelessness (UNVOI), since they are reported to indicate the lenition of the phoneme. To perform the task, we retrieved the acoustic parameters’ values. Next, we excerpted randomly 30% of the tokens and classified them perceptually as either retained or weakened. Statistical analysis revealed a weak, despite significant, correlation between the parameters, but a satisfactory correlation between DUR/COG/UNVOI and the results of the perceptual test, which is a good prediction for the elaboration of a trained algorithm for classifying the fricatives in big datasets of non-laboratory speech samples.