My project finds the 25 most frequently said words in a select set of presidential speeches from AmericanRhetoric.com, and compiles the results in a video format. It shows the respective presidents saying those words. I wanted to create something related to sounds and voices, at the beginning of my process, and explored what are some interesting ways to use a forced aligner. Since a forced aligner lets me know what words are spoken exactly when in an audio, I thought it would be worth looking at how different presidents say the same words. I looked at the word and phrases containing “God,” “believe,” “men,” “women,” and discovered that the way they say them are quite similar–i.e. they pause after “God” and enunciate it strongly (see below video). I wanted to incorporate these observations into an informative video, so I decided to string together the top 25 most frequently said words in presidential speeches.
The method I used to create this is straightforward. First, I used BeautifulSoup.py to scrape presidential speeches and their videos from AmericanRhetoric.com, only using results that contain a video and that are spoken by presidents. Now that I have the transcripts and videos, I used the Gentle Forced Aligner to process each speech, extracting the time when each word is spoken. The results are often accurate, but sometimes the aligner is not able to process some parts of a speech, which I realized later on. So now I have transcript, video, and JSON files of when words are said in the transcript for each video. I then wrote several Python scripts that helped me find the most common words in these speeches, phrases containing these words, and sentences containing these words. I tried different ways to use the information I have but finally settled on just using the words themselves, not the containing sentence or phrase, for this project. Once I have the list of words I need with their respective timings, I used Movie.py to string together clips from each video in a sequence that I needed, and to add scaffolding, i.e. number and word for each section of a word.
I could improve this by using more sources and videos, right now I am only using 20 but with more videos the effect of it might be more impressive. My program could also be used to show some interesting things, like how each succeeding president regarded the previous one. With today’s presidential campaigns I could also apply the same analysis of most commonly said words to each candidate, and create videos for each of them.