How has music changed over the past 100 years?

It’s fairly obvious that music has changed over the past 100 years - as anything naturally does over such a long period of time. However, an interesting analysis can be done to explain how music has changed over the years. Possible changes of interest are duration of songs, explicitness of songs, mood of songs, and certain characteristics of a song such as tempo, energy, instrumentalness and loudness - among others.

This data set contains 160,000 tracks from spotify with the following data. Some of the variables are defined below.

A Few Caveats

While this data is extensive, it is by no means complete. There was far more than 160,000 tracks that were recorded and released during this 100 years. This data is simply a sample of tracks released in those 100 years. The data is also compiled into yearly averages which is what this analysis uses - for the most part.

Moreover, the sample of tracks in this data set are just that - recorded tracks. This does not necessarily mean the tracks are music - as some of the variables hint at. They could be music, talk radio, podcasts, or news shows.

Finally, some of these metrics are bound to favor more recent tracks compared to older tracks. For instance, popularity.

  • Popularity: Numerical, the popularity of a track is a value between 0 and 100, with 100 being the most popular. The popularity is calculated by algorithm and is based, in the most part, on the total number of plays the track has had and how recent those plays are.

Under this definition, this only counts todays downloads/plays. It is likely true that the average spotify user skews to the younger side. This skew would likely mean the average spotify user is player music that has been created more recently rather than 100 years ago. Thus, while the popularity measure may be accurate for more recent tracks, it is unlikely accurate for older tracks. Further, popularity of more recent tracks will be higher simply because they are newer and recently released. Older music popularity was not calculated when it was originally released because downloads and plays were not metrics that could be calculated - thus they had to be back calculated with today’s plays of old music. For the reasons listed above, varaibles that tend to favor new music were not included in this analysis.

Please note that this analysis may use the word “music” and “track” interchangably - though it is known that the data set is not exclusively music.

Denisty Plots for song length

I split the data into 4 distinct sections of about 25 years. This will help give snapshots of the distribution (spread) of duration of songs throughout history.

Overall, the distribution of the 25 year splits was not too telling. From 21-45, you see a peak density at 3.15 minutes. From 45-69 there is a peak density at 2.8. From 70-94 there is a peak density of 3.71 and finally from 95 - 20 you see a peak density of 3.14. Interestingly you see an ebb and flow to the 4 snapshots, increasing and decreasing throughout time. This generally makes sense as there are few trends that go in one direction, unquestionably, for 100 years.

What might be more telling is a year by year density plot. The animation below shows the density of track duration for every year. What you can see, interestingly, is not that songs have been decreasing in length since 1921, but that there is an actual decrease in song length until 1940(ish), and increase in song length until 2000(ish) and finally a decrease in song length afterwords. This confirms what we say in the four density snapshot.

Danceability and Instrumentalness

The first metrics I examined where danceability and Instrumentalness. danceability is an interesting metric and seemingly difficult to quantify. In this case, it is a combination of musical elements including tempo, rhythm stability, beat strength, and overall regularity that make a song “danceable” Similarly, instrumentalness is a numerical figure that predicts whether or not a track contains no vocals. It is important to note that instrumentalness is a predictive value. A 1 indicates that the song likely has vocals while a 0 indicates the song is most likely instrumental. It is not how much of the song is an instrumental. The graph below shows the yearly breakdown of both danceability and instrumentallness. In this case, insturmentalness is interpreted as the probability of a song being instrumental while danceability is the mean “danceability” of songs that year.

Separately, danceability has increased over the past 100 years, with year to year fluctuations. On the other hand, instrumentalness - or the probability of a song having no vocals - has fallen year over year.

The correlation between these two variables is a fairly strong -0.49.

Danceability and Energy

Below is a graph between danceability and Energy on a year by year basis. The definition of energy below:

  • Energy: Energy is a measure from 0.0 to 1.0 and represents a perceptual measure of intensity and activity. Typically, energetic tracks feel fast, loud, and noisy. For example, death metal has high energy, while a Bach prelude scores low on the scale. Perceptual features contributing to this attribute include dynamic range, perceived loudness, timbre, onset rate, and general entropy.
    It tracks that as energy increases, so will danceability. Generally, people won’t dance to Bach but will to rap.

The above graph shows the relatively strong correlation between Energy and Danceability. Generally, you see a downtrend in both danceability and Energy between 1921 and 1950. Then, a sharp increase in Energy after 1950. With that sharp increase in Energy, there is a corresponding increase in Danceability.

Remaining variables

Finally, we will take a look at acousticness, loudness, speechiness, and valence.

Acousticness

Overall, on the messy graph you see above, Acousticness - which is measured as the confidence on whether or not a track is acoustic - is fairly high up until 1950 when it falls off a cliff. This is a fairly stark change in track acousticness. Generally, I think this can be attributed to new technology and innovations. First, in 1940 the Covoder was invented. This was an early form of auto tune. In 1949, the 45-rpm Record was invented. These provided crisper, clearer sounds and noise reduction. Similarly, more advanced microphones were invented just after 1951 along with 10 cheaper versions of electric guitars. Finally, the Transistor Radio was invented in 1954. This allowed for more widespread consumption of music. Overall, acousticness decline was largely due to both technology inventions and innovations which allowed for the clearer recording of music through noise reduction - allowing for less tracks to be acoustic.

What was the outcome of this technology change?

While it may have been fueled by the changes in technology, rock and roll “officially” started - according to wikipedia - in 1950. Generally, Rock ‘n’ Roll is not considered to be an acoustic form of music.

Loudness

The line graph of loudness is below. Loudness is measured by decibels. The closer to 0, the louder the sound.

Loudness, perhaps unsurprisingly, has steadily increased over the past 100 years. While music has become louder, it has also become more consistent. This increase in loudness begins in 1950 - a year we have mentioned in this analysis - the year Rock ‘n’ roll started.

Speechiness

Below is the “speechiness” of tracks. The definition of speechiness is below.

  • Speechiness: Numerical, speechiness detects the presence of spoken words in a track. The more exclusively speech-like the recording (e.g. talk show, audio book, poetry), the closer to 1.0 the attribute value. Values above 0.66 describe tracks that are probably made entirely of spoken words. Values between 0.33 and 0.66 describe tracks that may contain both music and speech, either in sections or layered, including such cases as rap music. Values below 0.33 most likely represent music and other non-speech-like tracks.

The graph for speechiness is year over year - meaning the values for each year are an average speechiness of tracks in that year.

Essentially, this graph shows the average “speechiness” of tracks pulled from those years. A higher number means there are more tracks that are talking tracks (books on tape, talkshows, podcast etc.,) while a lower number means the tracks are mostly music. Generally, what you see is a downtrend in speechiness (or more music) until 1975 and then an uptrend in speechiness (or less music). Perhaps most notably is the uptick in speechy tracks after 2005(ish). This may be the due to the increase in podcast/news shows on music streaming services.

Valence

Valence is defined as the general positiveness conveyed by a track. 1 means tracks are largely positive in a year while 0 would mean tracks are largely negative.

In this case, positive could be happy or cheerful while negative could be sad, depressed or angry.

Interestingly, but perhaps unsurprisingly, music has been in a downtrend in positivity over the last 100 years. This downtrend reached its low just recently in 2017 - assuming that prior to just 1950 may not have the best data available. While we saw a fairly strong bounce from this low, music is still fairly negative, overall.

Conclusion

Overall, music has changed in the last 100 years - which is unsurprising. What surprised me most, however, was the lack of change in song duration. While there was general fluctuation, songs now vs. in 1921 are pretty much the same length.

What is more interesting are the great number of changes that occurred in music over the last 100 years - because pretty much everything changed. This change largely occurred because of the introduction of Rock ’n Roll. Since 1950, you see an increase in Danceability and subsequent decrease in instrumentalness. Furthermore, just after 1950 you see a sharp increase in track “energy.” This same change is after 1950 is apparent in track acousticness (decrease) and track loudness (increase).

Furthermore the popularity of talk tracks, such as talk radio and podcast, are on the rise in recent years.

Finally, track’s “positivity” have been trending downwards since 1925. While I would be surprised if this trend will change over the next few years, there has been a drastic bound in track positivity since 2017.