Audio Alone Can Spot Sports Highlights, Study Shows

Most people think watching sports highlights means focusing on what you see. But a new study asks a bold question: can sound alone pick out the best moments from a game? Researchers tested this idea using only audio from sports videos. They built simple models that listen instead of watch. These models used pre-trained audio tools to understand sounds like cheers, whistles, and commentator voices.

The team compared their audio-only models with methods that use video or both audio and video. They tested everything on a standard sports highlight dataset. Surprisingly, their basic audio model did better than some older methods that used both sight and sound. This shows that listening can be powerful on its own.

When they combined audio and video together, the results got even better. This means seeing and hearing give different clues. Neither one alone is enough. The best performance came from using both at the same time.

To dig deeper, the researchers split the audio into parts. They separated crowd noise from commentary. They found that the commentator's voice gave more useful info than just background sounds. But mixing both types of audio worked best overall.

They also looked at specific sound patterns. Highlight clips had louder volume and more energy in mid-range frequencies. But there was too much overlap between highlight and non-highlight sounds. So volume alone cannot tell the full story. The study suggests audio should not be an afterthought in highlight detection. It deserves a leading role.