Quantifying the Diva-ness of 138 National Anthem Performances – Beragampengetahuan

Now that you know what to look and listen for in the “Star Spangled Banner,” it’s time to dive even further into the data. The heatmap below allows you to explore each of the 16 phrases for all 138 performances in our dataset. Surprisingly we didn’t get tired of listened to the anthem when doing this project, so maybe you won’t either!

If you want to hear more about our favorites, check us out on the Switched on Pop podcast.

Methods

To build a catalog of celebrity National Anthem performances, we consulted several best-of lists, and collected performances from major sporting and political events like the Super Bowl, the World Series, and the presidential inauguration from 1980 on. We also included a few notable anthem performances prior to 1980.

We then set out to find a suitable audio/video recording of the performance on YouTube. We were able to find roughly 180 performances. Using a free online converter, we extracted mp3 audio files from each YouTube link. We then manually trimmed the mp3 audio files so that they started right before the first word in the anthem and ended after the last. The trimmed audio was then uploaded to an AI-powered Vocal Remover program to isolate the vocals from the rest of the audio.

Once we had the isolated vocal files, we loaded them into Tony, an open source melody transcription tool developed by the Centre for Digital Music at Queen Mary, University of London. Along with 4 additional data assistants (Amel Awadelkarim, Dustin D’Andrea, Noah Fagan, and George McIntire), we listened to each vocal track, manually removed errant sounds (crowd noise, echos, jet flyovers, band accompaniments) and corrected pitch when necessary. Some performances were dropped at this stage due to poor audio quality.

We then exported the pitch data, which tracked frequency in Hz over time. We manually marked the starting and ending timestamps of each of the 16 phrases of the song. Some specific phrases for specific performances were also dropped at this stage due to poor audio quality. This process took over 100 hours of work and we couldn’t have done it without our data assistants, so thank you!

For each phrase, we used Dynamic Time Warping (DTW) to compare it to the corresponding phrase in a “standard” performance of the National Anthem. DTW measures the similarity between two temporal sequences, neutralizing any variation in speed. In other words, it measures the amount a performer deviates from the standard, strictly melodically. This is what we called the Diva Score for each phrase. The Diva Score for a full performance was calculated by taking the average for all phrases from that performance. It’s not an exact science, but we think the Diva Scores match our intuition of what melodic Diva-ness is pretty closely!

Finally, thank you to our colleague Kevin Litman-Narvarro who originally planted the idea of looking into National Anthem performances.