 I'm going to try to freestyle this. All right, so quick disclaimer. All the video that you see in this presentation, I shot on my phone. All the video that you don't see in lieu of me getting permission to use it, I want you to just use your imagination. All right. I'd like to talk to you about Johnny Cash. If you're not familiar with Johnny Cash and his work, he was born in 1932. And he grew up in Dias, Arkansas, picking cotton on his family's farm. And then he started his music career in 1954 and gained nationwide popularity with a string of early hits. And then he went on to record across five decades and in multiple genres. He was inducted into the Country Music Hall of Fame, the Rock and Roll Hall of Fame, and the Gospel Hall of Fame before. Finally he passed away in 2003. And this year marks the 50th anniversary of the recording of At Folsom Prison, which is an album he made in Folsom Prison. And his label Columbia at the time was reluctant to release such an album. But they released it and it was a huge success for Johnny Cash. And is a good example of Johnny Cash challenging the conventional boundaries of the music industry. Were you there? Now another example of challenging convention, Johnny Cash initially wanted to record gospel music and he established a pattern when he felt reluctance from record companies. He would have a hit song or hit album and then his next project would be purely a gospel project. And it wasn't until he started recording with Rick Rubin as his producer that he enjoyed some freedom and he was able to record gospel without a producer worrying about balancing the sensibilities of a non-gospel and a gospel audience. There's a man going round taking names. So this song, The Man Comes Around, this is a especially vivid example of conviction and vision for a modern gospel. It represents an expanded rock sensibility. It's also a bridge to Johnny Cash's past. So we're going to work toward creating a visualization that measures the audible distance between The Man Comes Around and music at either ends of Johnny Cash's career. So what we need to do is identify Johnny Cash's gospel music in this enormous catalog of his recorded music. And you might think, OK, well, this is really easy. We'll just take all the tracks from his gospel albums. But if we do that, then we're going to miss important gospel songs that only appear on his mainstream albums. So we're going to look at his lyrics, and we're going to use a technique called topic modeling. Topic modeling is an algorithmic approach defining latent structures occurring in a collection of documents. So documents, in our case, are songs, and the topics are a list of words occurring in statistically meaningful ways across the collection of songs. So this approach assumes that words referring to similar subjects appear in the same context. And you can imagine an author constructing a body of text by pulling from buckets of words. And these buckets of words, maybe one bucket might be referring to human senses and another to trains. But when the author has laid out the core meaning of what they're trying to convey, then they can add some structure and you have a document. So topic modeling attempts to reverse engineer this process. We want to go from the songs to the buckets to this list of words that represent dominant themes in Johnny Cash music. So I created a topic modeling using some software and it uses 10 buckets. And the bucket that we want to use and focus on is the one that has words like God and word and Jesus. And it's going to be the topic that we're going to use to identify religious-oriented music in his catalogue. So what we're going to do first is represent every Johnny Cash recording as a single white pixel along this horizontal line and we're going to lay it out in sequential order by initial recording date in the track position. And when we map these songs, well, first of all, if we do this, we're going to have his earliest recordings, the recordings he made with Sam Phillips, his first producer at Song Records. And then at the right-hand side, we're going to have the last recording he ever made and this was with Rick Rubin. And we're going to map all the songs that we've labeled as religious-oriented songs and you're going to see an interesting distribution. There's a cluster at the left at the very beginning, a cluster in the middle and a cluster at the right. So we're interested in comparing early to late. So we're going to filter out everything in the middle and we're going to get 16 explicitly gospel songs. And this leaves us with 16 songs. It's nine in the early gospel period. This is Song Records, Columbia, and then seven that are in the late gospel period and these are all exclusively with the Americans Recordings Projects and Rick Rubin. All right, you may recognize this as a waveform. So now we've got our gospel songs. We need a way to compare them and we need to do that audibly, right? So this waveform, we usually associate these with digital recordings and the oscillations and amplitude represent a change in input signal over time. And looking at this, we can infer certain things about the sound. We can infer length or possibly tempo, but if we want to get more detail, we need to use something called spectrum analysis. So spectrum analysis, it renders volume levels of specific frequency ranges. So this represents a tenth of a second on an open A string on a violin and those peaks that you see at the sixth octave register, those represent the most dominant features in the sound. So if we combine spectrum analysis with machine learning, then we can derive higher level audio attributes, right? So dance ability and acousticness and loudness, and then if we combine those, we can get an even higher representation of the overall audio attributes called an acoustic vector. All right, so now we can represent a song as a vector, but what is a vector? It's a really quick. A vector represents a physical quantity that has both magnitude and a direction. That's a lot of math really quick. But what's really nice is we can use vector math to add and subtract vectors, and we can even then get the average of two or more vectors, and we're going to call that the mean vector. And we'll use the mean vector when we're comparing periods early and late. So we're going to build a really quick example, a toy example. We have a list of songs, and let's say that we measure the level of treble and a level of bass in these songs. We map those levels between negative one and one. I can represent all of these songs as an array of two-dimensional vectors, but then I can plot them, right? So now we can get a sense of how they relate to each other, and I can take then the average vector and do some interesting comparisons. So this song that is closest to the average, it sounds the most similar. It sounds most like the average, while this song is the furthest away, and it sounds the most distinctive. All right, so when talking about the man comes around with Robert Hilburn in an interview, he said, Johnny Cash said, if someone is still listening to my music 50 years from now, if someone is listening at all, I hope they're listening to that song. So here he is. He's written hundreds of songs where at the end of his career, and he's telling us, he's written his most important song. So it's interesting because in the same interview, he tells us explicitly that this is a gospel song, but it doesn't sound like the same gospel songs that were written in that later period, but this was intentional. And if you hear, you can hear that kind of strumming on the guitar. That was something like a signature sound and Marty Stewart, the guitar player on this track, actually played Luther Perkins' Fender Esquire guitar that was used in those earlier recordings. So it went way out of their way to sound authentic, but at a much later period. So to visualize the difference between the man comes around in the late gospel recordings, so let's say this dot represents the acoustic meaning of all the late gospel songs, plus the man comes around, and the radius represents the audible distance from the acoustic average. The most similar sounding song in this collection is Redemption of American Recordings. And then you can guess that this one out here is the man comes around. So this is the most distinctive of the group. And then we can compare the relative magnitude of the man comes around to the acoustic mean of both the early and late gospel. And you can see it's pretty distant from both. And then if you flip this over and knowing what we know about the intent of the song, it was explicitly meant to reference this earlier sound, we can kind of think of it metaphorically as this fulcrum between the early and the late gospel periods. So the American Recordings period, it marks the first time where Johnny Cash felt free and uninhibited to record gospel. And it kind of invites us to compare all the early gospel to all the late gospel instead of just between the man comes around. So we're going to fuse these two periods together and we're going to put the late gospel on the right and again the distance from the half screen there represents the distance, the acoustic mean of that period. And then we can compare to the early gospel and you can see that the late gospel, the songs are more distinctive and they cover a broader sonic range. Okay, the question here now is why? Well, there's a darker side to Johnny Cash's gospel. It's not all good news. In Sam Phillips' audience wasn't ready for good news gospel, uh-oh, or bad news gospel, but Rick Rubin's audience was fully ready to accept more realistic lyrical themes and fully accepted Johnny Cash in his music. So the difference in attitude between these two producers represents modern rock and roll's expanding sensibility. Early rock was narrowly defined and intentionally segregated from other genres. But by now modern rock audiences have developed a sophisticated ear and broaden the genre's boundaries to include more diverse references. So one of these is gospel and when you mix modern rock with, you know, gospel and especially kind of the darker interpretation, the darker side, you get a mix that's appealing to a modern rock audience. So listen to the audience right here. Whoops, sorry guys. Bad timing. Can we turn the, uh... So I don't know, you know, how often everyone in this in this venue at this concert goes to church, but they all know every word of the song and you'll hear it. And I bet most of you know the words as well. So thanks.