Exploring Temporal Reduction In Dialectal Spanish: A Large-Scale Study Of Lenition Of Voiced Stops And Coda-S

Large scale studies of temporal reduction are of interest to obtain linguistic knowledge and this understanding can potentially help improve speech technologies. Automatic speech recognition (ASR) supports the processing of very large corpora and can be used to enrich linguistic studies with models of speech production and perception grounded on observed commonly used pronunciations. In return, ASR can benefit from the linguistic findings by including pronunciation variants that reflect the linguistic variability. This study focuses on two temporal reduction phenomena in Peninsular and Latin American varieties of Spanish: lenition of intervocalic voiced stops V/bdg/N and coda-s. First, the two phenomena are investigated via a study of transcription errors produced by a speech recognition system designed for Peninsular Spanish which can be potentially attributed to lenition. In a second step, automatic forced alignment experiments are conducted using specific pronunciation variants with and without lenition to measure the extent of the phenomenon as a function of geographical and stylistic repartition. The results show that the distribution of pronunciation variants across Peninsular and Latin American Spanish varieties is consistent with trends depicted by classical linguistic studies. The speaking style appears to be the main factor affecting the +/-lenition variation. The findings also suggest that including such variants in ASR system's lexicon may improve performance when processing multiple Spanish varieties.