Hello, I have 26 hours of audio data. This is huge data to manually label every 2s data frame as a scream or not. Is there any idea to do at least a portion of it automatically to save time?
Dear Sree Nirmillo Biswash Tushar,
I suggest you to see links and attached files on topic.
http://dcase.community/challenge2021/task-sound-event-detection-and-separation-in-domestic-environments
https://docs.microsoft.com/en-us/azure/machine-learning/how-to-create-labeling-projects
http://tprn.reflectivemanagementacademy.it/keyword-extractor-from-site.html
Best regards
Sree Nirmillo Biswash Tushar ,
You should calculate the decibel levels on a sliding window to cover the 2second interval that you are using. From there you can put a threshold on what is considered scream or not (Using a standard classification of noise. For example classify screams at night as noise falling in the 50 dB as per the WHO's NIGHT NOISE GUIDELINES FOR EUROPE or similar guideline).
Thank you very much. A baseline will reduce the amount of search.
Dear Sree,
You can write a program to iterate through the audio data every 2 sec. to search for a threshold to automate a label for a scream. You can use GPU hardware to speed up the search.
- Badges
- Science topic
- Similar topics
- Filing