It’s a great first approach to the issue, however, there are some limits and difficulties (if I understood this approach correctly that is):

1. The approach relies heavily on a pre-existing corpus of 14,000 pop songs spanning back to 1950. This means that the system may not be able to accurately identify similarities in songs that were not included in the corpus or that were released after 1950.
2. The approach utilizes a threshold of 0.24 to determine if there is an infringement. This threshold may not be appropriate for all cases and may lead to false positive or false negative results. Of course, we’ll need more testing before we see start assessing all musical claims.
3. The approach only compares melodic lines and doesn’t take into account other factors such as lyrics, chord progressions, and timbre. This means that the system may not be able to accurately identify similarities in songs that have different melodies but share other elements.
4. The system’s 90% accuracy in recreating the outcomes of historical copyright cases is based on small sample size and may not be representative of all cases.
5. The system is not a legally binding decision and the final decision will still be left up to the court to decide. But again, this is just a matter of time.

As with any other approach, further research and testing are necessary to improve and validate the system’s accuracy and reliability. But it looks promising!