How does the brain encode speech sounds?
Integrating phonological theory with neurophysiological ontologies, we investigated how the brain encodes the speech acoustic signal into phonological representations. Previous studies failed to show clear neural hints for speech processing. Using high-density electroencephalography, we mapped the Salento Italian vowel system onto cortical sources: our findings showed for the first time that primary and secondary auditory cortices are characterized by a hierarchical processing that dynamically transforms the acoustic signal into phonological representations according to the distinctive features principle. These findings give definitive evidence in favor of the classical theory developed in the ‘50s by Romans Jakobson, Morris Halle, and Kenneth Stevens: the Analysis by Synthesis Model.
How do learners acquire L2 speech sounds?
Our longitudinal behavioral and neurophysiological investigations on the acquisition of second language phonetics and phonology in formal context (classroom), demonstrated that discrimination and pronunciation of L2 sounds do not improve at all. This is due to the poor quality and quantity of L2 stimuli received by learners, with teachers have a non-native accent, and students converse with each other in their first language during activities. These results were later confirmed by other independent laboratories with different L2 learners. Along this line, our research is demonstrating how focused trainings can reactivate auditory and articulatory plasticity.
Recently, we showed that visually prominent articulatory information did not boost auditory identification of L2 sounds in adults (despite observed benefits of audiovisual integration during the training sessions). For adult learners, the process of acquiring L2 phonological segments begins with analyzing the acoustic signal and attempting to map its properties onto long-term memory representations, such as distinctive features. Distinctive features contrastively active in the L2 system but not contrastive in the L1 system constrain constrain the learning process.
How can we characterize phonetic-phonological variation?
Investigating dialects variation in southern Salentino varieties, we discovered phonological processes (metaphony) that, according to the classical literature, should not have been present in this linguistic area. Thanks to acoustical and statistical analysis of data, we revealed the presence of metaphonic raising both for stressed /ɛ/→[e] and /ɔ/→[o] when followed by unstressed [i] or [ʊ]. Crucially, we showed that this phenomenon is interested by extreme variation. These findings have posed critical challenges both for phonological theories and historical linguistics, suggesting new ways to look at dialectal variation. Further articulatory and neurophysiological data confirmed the initial findings and showed that the distinctive feature [±ATR] is involved in the raising of mid-vowel.
How can we compare the voice for forensic purposes?
We tested different variants of a Forensic Automatic Speaker Recognition (FASR) system based on Emphasized Channel Attention, Propagation and Aggregation in Time Delay Neural Network (ECAPA-TDNN). To this scope, conditions reflecting those of a real forensic voice comparison case have been taken into consideration according to the forensic_eval_01 evaluation campaign settings. Using this recent neural model as an embedding extraction block, various normalization strategies at the level of embeddings and scores allowed us to observe the variations in system performance in terms of discriminating power, accuracy and precision metrics. Our findings suggest that the ECAPA-TDNN can be successfully used as a base component of a FASR system, managing to surpass the previous state of the art, at least in the context of the considered operating conditions.
Our software is currently one of the most reliable options available for forensic voice comparison.
© 2006-2025 Centro di Ricerca Interdisciplinare sul Linguaggio
