BrSD is a novel dataset freely available created to support the development of speech-based recognition tasks. It is composed by 400 utterances from 80 different contributors, all of them brazilians and portuguese speakers. It is recommended for classification tasks related to speech recognition, for instance age, gender and accent.
The database was developed by Marco Aurélio Deoldoto Paulino, Alisson Renan Svaigen and Yandre Maldonado e Gomes da Costa from State University of Maringá (UEM), Maringá, Paraná, Brazil.
BrSD had the following contributors in its development: Linnyer Beatrys Ruiz Aylon, from State University of Maringá (UEM); Alceu S. Britto Jr., from Pontifical Catholic University of Paraná (PUC-PR); and Luiz E. S. Oliveira, from Federal University of Paraná (UFPR).
This website provides to researchers the following items:
A comprehensive specification of BrSD structure, speakers, utterances and studies carried out is available here.
You can download the BrSD version 1.0 here
The zip file is organized as follows:
Spectrograms are a visual representation of the audio signal and have been widely used in many different audio classification tasks.
We provided spectrograms from 400 BrSD utterances. They are excerpted with SoX, a free software that can convert various computer audio file formats into others. The generated spectrograms have frequency limit of 3.4 kHz and the amplitude range from -60 dBFS to 0 dBFS.
Spectrograms can be downloaded here.
We provided four acoustic-based features and one visual-based feature for researchers. The available features are:
RP, RH and SSD was generated by RP Extract framework in its standard parametrization.
Features can be downloaded here.
Paulino, M. A et al. A Brazilian Speech Database. In: Tools with Artificial Intelligence (ICTAI), 2018 IEEE 30th International Conference on. IEEE, 2018.
For suggestions, contributions or direct contact with creators, you can send an e-mail to: