Broadcasters produce and transmit a vast number of sports videos in cyberspace due to immense viewership and potential commercial benefits. The analysis and processing of such a huge amount of video content are very challenging. This situation demands the development of effective and efficient summarization methods to manage the massive sports video repository while keeping the viewer’s interest along with potential storage and transmission benefits. This paper presents an automated summarization framework based on excitement detection for sports videos i.e., cricket, soccer, etc. The audio stream of the sports video is analyzed to capture the significant events that are then used to produce the concise video. For effective representation of audio signals, we proposed an acoustic feature descriptor symmetric ternary codes and used them to train a binary Support Vector Machine classifier for excitement detection. Each audio frame is labeled as either an excited audio frame or a non-excited audio frame. The video frames corresponding to the excited audio frames represent the key-events in the sports videos and are marked as the key-frames. Each key-frame is appended with the neighboring frames to produce video skims for each key-event based on the user’s required summary length. Finally, these video skims are sequentially arranged to produce the user-driven video summary. We evaluated our highlights generation method on our own diverse YouTube dataset of cricket and soccer videos, and a largescale SoccerNet corpus of soccer videos. The average accuracy of 97.7% and 91.23% on both datasets confirms the reliability of our method in terms of key-event detection for sports highlight generation.