A speech database for the study of Lombard effect in Brazilian Portuguese

The Lombard effect is related to the change in the way people speak in noisy environments in order to compensate the degradation of the audio signal. There are significant changes in the duration of some types of phones and in the formant positions that can vary from a language to another. The construction of a speech database that isolates the Lombard effect is a challenging task because its necessary to record only the speech signal, without the environment noise. In this paper a methodology to construct such databases is described and a database for the Brazilian Portuguese was produced using this procedure