MALTA

MALTAav is a dataset consisting of 492 videos, with an average length of 126 seconds and around 5 sentences describing every video in each of two languages, viz., Marathi and English, along with background speech in Marathi rich in content.

The ground truth of MALTAav was generated by instructing close to 10 annotators to pay close attention to the audio as well as the visual streams while aligning the sentence captions with the video.