| Summary: | ABSTRACT : Audio analysis is a topic of study that has gained momentum in the last decade, the growing information as well as the improvement in computational power has allowed more and more academic and industrial sectors to perform studies of audio signals which previously went unnoticed. With this type of analysis certain drawbacks arise, one of them is that in many cases the recording conditions will not be optimal to obtain a sample with "clean" information, because external factors affect or introduce noise to the sample. As a solution to this problem, multiple algorithms have been developed for audio cleaning, some of them require manual work that can be exhausting depending on the size and quantity of audios, and on the other hand there are techniques that use predictive models created with Machine or Deep Learning to perform the cleaning process in an automated way. Although these last techniques have solved the problem of doing this work manually, many of them are not user-friendly and require the user to have knowledge of the model created in order to make changes and experiment at ease, thus reducing the number of people who can make use of this technology. In this work a web application was created which allows to make use of a Deep Learning model called ORCA-CLEAN [23], created to perform audio cleaning for whales. and couple it in such a way that the user can perform audio cleaning without having knowledge of the model and just making use of his mouse and keyboard. The user can select multiple regions in the audio spectrogram in order to apply different types of parameters and make comparisons, as well as listen to the resulting audio(s) after applying the cleaning process. Finally, the user can download a zip folder containing images of the spectrograms of the regions before and after cleaning, as well as the cleaned audio(s).
|
|---|