From September 18, 2024 to December 18, 2024
Ana Montalvo, from CENATAV, was hosted for 3 months by Avignon Université
As part of the Esperanto project, she worked for three months at LIA on the topic of pathological speech with Corinne Fredouille and her team. Based on the arc
hitecture used by their system for Pathological Speech Quality Assessment, she designed, implemented and evaluated multitask models to simultaneously estimate intelligibility and severity. She assessed several non-intrusive metrics for Speech Quality Assessment and compared them with the estimates from the multitask model. They are currently working on the model's explainability and the interpretability of the results.
As part of the periodic meetings of LIA's Speech and Language Group, she gave a talk on applying non-semantic task learning to improve spoken language recognition in very short-duration signals.
During the last month of her stay, she worked on topics related to speech synthesis, with a particular focus on modeling dialectal accents of Latin American Spanish. She participated in a one-week exchange with colleagues from LIUM, where she gave a presentation on the challenges of TTS modeling for Cuban Spanish, laying the groundwork for JSALT 2025.
