In 2022, a Swiss client approached us with an interesting task: to label audio recordings from microphones attached to sheep shepherding dogs. The microphones would activate upon detecting noise, and the client needed each sound categorized into predefined categories such as barking, howling, growling, or no noise at all. The challenge was to process and label around 13,000 audio recordings, each lasting ~5 seconds while filtering out irrelevant background noise.
 

Year: 2022
 

Tool Used: Google Spreadsheets for labeling output
 

Challenges We Faced:
The audio recordings, collected in various outdoor environments, contained a mix of noises, including dog sounds and ambient background noise, making the task more complex. We had to ensure that the correct sound—barking, howling, growling, or silence—was accurately identified and labeled from each 5-second clip. Additionally, the volume of data required a quick turnaround, with the client requesting the work be completed within a few days.
 

Despite the time constraints and the challenge of filtering out irrelevant noise, Abelling successfully delivered the project. Our annotators used precision and speed to label all audio clips with the highest accuracy, ensuring the client received a dataset that could be efficiently used to train their models.
 

Just another successful project demonstrating Abelling’s capability to handle unique data labeling challenges!