Finished! Looks like this project is out of data at the moment!
Results
Data collected thus far
Thank you to all of the annotators that have annotated our urban soundscapes thus far!
Below is a graph of the distribution of the labels we have collected thus far. The length of each bar represents how many labels were reported for each class, where each class is counted at most once per recording. From this chart, we see that "Person or small group talking" was annotated the most, followed by "Large-sounding engine" and "Medium-sounding engine". Whereas "Ice cream truck" was annotated the fewest number of times. Through this annotation campaign, we are hoping to get 1000 example positive labels for each class.
Up to three annotators annotated each recording. Therefore, we have also broken down the number of labels by the how many of three annotators reported the label. For example, if only one of the three annotators reported a "siren" in a particular recording, then that label contribution would be included in the blue segment of the bar. If all three annotators reported a "siren" in the recording, then that label contribution would be included in the green segment of the bar. In the next figure, we normalize the lengths of the bars to all be the same so that we can more easily see the distribution of label agreement for each class. From this chart, we see annotators more often agree on easy to identify classes such as "Dog barking/whining" and "Siren" and less often on the classes "Amplified speech" and Mobile music", which may be difficult to correctly identify.
