Constructing a Dataset of Speech Recordings with Lombard Effect

摘要

The purpose of the recordings was to create a speech corpus based on the ISLE dataset, extended with video and Lombard speech. Selected from a set of 165 sentences, 10, evaluated as having the highest possibility to occur in the context of the Lombard effect, were repeated in the presence of the so-called babble speech to obtain Lombard speech features. Altogether, 15 speakers were recorded, and speech parameters were calculated and analyzed. First, a brief summary of the research related to the Lombard effect is given. Then, the recording studio characteristics and the equipment utilized for recordings are shown. Examples of analyses are included, concerning both non-Lombard and Lombard speech. Finally, a recapitulation of experiments performed along with further research plans is given. The link to the data is also provided.