Pré-processamento de sinais acústicos para análise e caracterização de vocalizações de lobos-guará

The possibility of automatic species identification through their communication acoustic signals opens new possibilities for the study of behavior and the preservation of biodiversity. Although commercially available, automated analysis tools still lack success in terms of outcome and accuracy. The success of an automated method requires the validation of a process sequencing strategy aiming at the determination of vocal acoustic parameters in the middle of a complex soundscape. A challenge in this regard lies in the extraction of noise from field recordings, which sometimes makes difficult the correct interpretation of acoustic parameters. In this study, we explore a combination of signal pre-processing steps to extract the noise of recordings of roar bark vocalizations of maned wolves (Chrysocyon brachyurus) in recordings made by a digital recorder sampled at 44.1 kHz with 16-bit resolution, using processing tools of the software Audiosculpt 3.2.7 by IRCAM. The procedure involved a sequence of steps: selection of the best audios and file segmentation; loudness normalization; identification and removal of noise zones. The results include processing over 10 hours of field recording with the production of a body of more than 131 vocalizations, under conditions suitable for the individual characterization analysis that will be done next.