HUPA: A corpus of disordered and normophonic voices in Castilian Spanish

Advances in the evaluation of voice quality and the automatic screening of voice disorders are currently constrained by the scarcity of large, publicly available datasets that include speakers of diverse native languages, which hampers the reproducibility and further development of research in this field. To mitigate this limitation, the present manuscript introduces the HUPA corpus, a comprehensive clinical dataset comprising sustained-vowel voice recordings from 440 native Castilian Spanish participants, including 239 healthy controls and 201 individuals diagnosed with a wide range of voice disorders. In addition to the audio recordings, the corpus provides subjective evaluations of voice quality performed by three expert raters using the GRBAS scale (Grade, Roughness, Breathiness, Asthenia, Strain), as well as a set of acoustic parameters that are automatically extracted from each voice record to offer an objective characterisation of vocal attributes. To the best of our knowledge, the HUPA corpus is the largest open-access, clinically validated voice-disorders database available of Castilian Spanish speakers. The HUPA corpus, therefore, constitutes a substantial resource for the investigation of the acoustic manifestations of voice disorders in Castilian Spanish participants and for the development and benchmarking of automatic voice pathology assessment methods. It has already supported multiple studies on automatic screening and acoustic feature design, illustrating its potential as a reference resource, while the broader challenge of developing robust, cross-corpus screening systems that generalise across languages, recording conditions and clinical populations remains an open research question.