VOICED Database

Abstract

This database includes clinically-verified 208 voice samples, from 150 pathological voices and 58 healthy voices. The database also includes information such as gender, age, pathology, lifestyle habits (e.g. smoking, alcohol and coffee consummation), occupational status, and the results of two specific medical questionnaires: the Voice Handicap Index (VHI) and Reflux Symptom Index (RSI).

Data Collection

Subjects involved in the study had to be adults aged between 18 and 70 years and able to follow several phases expected in the process. People aged under 18 and over 70, or with diseases such as folds or upper respiratory tract infections or with neurological disorders were excluded.

Healthy voices and the presence of vocal fold's disorders were clinically verified by the medical experts involved in the project. The database includes information such as gender, age, pathology, lifestyle habits (e.g. smoking, alcohol and coffee consummation), occupational status, and the results of two specific medical questionnaires: the Voice Handicap Index (VHI) and Reflux Symptom Index (RSI).

Medical phonatric examinations, voice signal acquisitions, and the completion of the medical questionnaires were performed at the ambulatories of Phoniatrics and Videolaryngoscopy of the Hospital University of Naples "Federico II", or at the medical room of the "Institute of High Performance Computing and Networking (ICAR-CNR)". The study started on May 16, 2016 and ended on May 15, 2017.

Voice recordings were made using an appropriate m-health system, Vox4Health, able to acquire in real time the voice signal by using the microphone of a mobile device. This system was installed on a Samsung Galaxy S4, Android version 5.0.1. It was held at a distance of about 20 cm from the patient at an angle of about 45 degrees.

All recordings were sampled at 8000 Hz and their resolution was 32-bit. Additionally, each recording was filltered with an appropriate filter to remove any noise accidentally added during the acquisition. The participants were instructed to articulate the vocal sample, with a constant voice intensity, as they would during a normal conversation. For each subject certain training tests were performed about two/three times before the recording.

Data Description

The voice recordings are provided in WFDB format and text format. In addition, the info text files describe the demographics of subjects.

The acquired signals consists of a recording of a vocalization of the vowel 'a' five seconds in length without any interruption of sound. All samples were recorded in a quiet (< 30 dB of background noise) and not too dry (humidity greater than 30-40%) room.

Ethics

All diagnoses were made according to indications of the SIFEL protocol, a clinical protocol compiled by the Italian Society of Phoniatrics and Logopaedics.

Acknowledgements

The VOICED (VOice ICar fEDerico II) Database was created by: Ugo Cesari (1), Giuseppe De Pietro (2), Elio Marciano (3), Ciro Niri (4), Giovanna Sannino (2), Laura Verde (5).

1. Department of Otorhinolaryngology, Universityy Hospital (Policlinico) Federico II of Naples, Via S.Pansini, 5 - Naples,Italy
2. Institute of High Performance Computing and Networking (ICAR-CNR), Via Pietro Castellino, 111-Naples, Italy.
3. Area of Audiology, Department of Neurosciences, Reproductive and Odontostomatological Sciences, University of Naples Federico II, Via S.Pansini, 5-Naples, Italy
4. Independent Doctor Surgeon Specialized in Audiology and Phoniatrics
5. Department of Engineering, University of Naples Parthenope, Centro Direzionale di Napoli, Isola C4 - Naples,Italy

Conflicts of Interest

The authors have no conflicts of interest to declare.