The secret recipe of hit songsFMP
Hit Song Science(HSS) is an emerging field that aims at predicting the commercial success of songs before they are released on the market. The term is coined by Mike McCready, an American entrepreneur in the music industry. He founded the Barcelona-based music analysis company Polyphonic HMI formerly, pioneering the use of an artificial intelligence tool to create significant value for music businesses. In 2003, the exploit of Hit Song Science was unveiled by Polyphonic HMI, they correctly predicted the success of Norah Jones’ debut album Come Away with Me months before it topped the charts, through the software that predicting the hit potential of music based on mathematical analysis of its characteristics.
The algorithm behind Polyphonic is a ”clustering” conception that locates acoustic similarities between songs (Elberse, 2005). When a new tune is fed into that computer, the particular algorithm called advanced spectral deconvolution works an astonishing dissection, analyzing the mathematical characteristics of music. Different sequences of Fourier transforms and mathematical functions break songs up, isolating tune’s patterns of rhythm, tempo, pitch, chord progression, the fullness of sound, sonic brilliance, and cadence. Instead of listening to the songs straight out, Polyphonic’s software constructs three-dimensional models underlying this data. The algorithm compares the new tune with the mathematical signatures of Top 40 hits in the last 30 years, then shows a kind of three-dimensional structure cloud filled in with dots representing songs. Inside the cluster, those hit songs tend to be grouped since they have the similar basis of structures. The closer the new tune is to the centre of one of those hit clusters, the more possible it get hit as well. The technology can be regarded as the reverse-engineering of music, and it is the way how they help record labels determine whether the song will have commercial success.
However, hit song prediction is still in its infancy and the accuracy is not satisfying so far. A larger-scale evaluation (Pachet and Roy, 2008) points out that for the popularity of music, some subjective labels (e.g. the “mood nostalgic” label) are analysed commonly by using the audio features, but the features are not informative enough to grasp anything related to such subjective aesthetic judgments. The negative results(Pachet, 2011) cast serious doubts on the predictive power of commercial Hit Song Science systems as well, it might not a reliable prediction approach because of the chaotic way individual preferences are mingled and propagated.
Nevertheless, Hit Song Science is still a wide-open field good to develop the music counselling businesses and study the dominant music style tendency. Hit potential equation (Ni, Santos-Rodriguez, Mcvicar, and De Bie, 2011), proposed by machine-learning engineers from the University of Bristol, analyzes the anatomy of the official UK top 40 singles chart over the past 50 years since 1961, intending to distinguish the most popular (peak position top five) songs from less popular singles (peak position 30 to 40) and quantify the hits feature tendency depends on the era.
This study looks at 23 separate musical features. The “w”s in the equation above (Walter, 2017) are “weights” or musical features including tempo, song duration, time signature, and loudness. They also compute more detailed summaries of the songs such as the “danceability”—how “danceable” the song is. These features are trained the publicly available data about songs in the UK chart, to prove its significance in producing a hit song. The “f”s in the equation means the exact same features of the proposed song which is compared with the past hit chords, and working out whether it correspond to the trends of the time. Finally, here comes a hit-prediction score, determining whether a song will make it to top five or if it will never reach above position 30 on the chart, with an accuracy rate of 60 percent.
Since the current popular musical style is ever-changing, the weights values have to be tweaked to match the era. In the 80’s, for example, slower musical styles (tempo 70-89 beats per minute), such as ballads, were more likely to become a hit. Plus, the danceability of a song got more relevant to its hit potential since the 80’s. And it was an interesting fact that hits around 1980 were particularly difficult to predict since the late seventies and early eighties were particularly creative and innovative periods of pop music. The varying dominant music style, culture and environment may lead the accuracy of hit potential equation varies over time. The contrast of this paper and previous studies is a possibly important qualitative difference, using the time-shifting perceptron to explain the evolving musical taste.
Shazam effect reveals more correlation between the current music trend and the preference of audience (Thompson, 2017). Shazam, as a mobile application for identifying unfamiliar songs through acoustic fingerprint for each track originally, pulls large amounts of musical data of people’s preference together. The data reflects an incredible real-time music popularity all around the world, knows which song is going to be hit before most of the people don’t know that. Now Shazam is utilized as a detection system for discovering the potential hit songs, and it definitely not the only source of data collection guiding the pop-music business. In fact, all of our searchings, streaming, downloading, and sharing on social media, iTune purchase and streaming music apps are tracking, this information is speaking out the relationship between music and the human mind, and the music industry is keeping find the answer: What do people want to hear next?
The result proves that human brains prefer melodies we already know (Huron, 2006). It estimates that when people listen to music, at least 90 percent of the time is spent on seeking out songs similar to what they have heard before. That’s because familiar songs are easier to process in the brain, and we prefer things that thinking with less effort, which is known as fluency in psychology. The power of habit makes people tend to listen to the similar music styles and genres again and again. In the past, experts in record labels and radios mastered the power that what kind of music the audience would listen, but data from the crowd have shifted the present balance of power. Music industry relies on the preference from audiences to raise the hit rate, to minimise their risk and even determine the future trend of music.
Will the dependence on data lead to the high similarity of styles and genres, promoting a dispiriting sameness in pop music? Not only are we hearing the same hits with greater frequency, but the hits themselves sound increasingly alike. According to the report released by Spanish National Research Council (Serrà, Corral, Boguñá, Haro and Arcos, 2012), pop music was growing increasingly bland, loud, and predictable indeed, while the diversity of note combinations has consistently diminished over the past half-century. The researchers concluded old songs could be made to sound “novel and fashionable” just by freshening up the instrumentation and increasing the average loudness, and all of it causes pop music gets more repetitive.
Trending music style doesn’t mean it always going to be popular, are there other ways that we can predict how a pop song affect us? It always hard to explain why we like our favourite pop songs so much, but we might link it to the biological response in neuroscience field. The recent study uncovers that harmonic surprise (Miles, Rosen and Grzywacz, 2017) in structure might be the secret to create a pop hit. In the majority of popular songs, the relatively rare chords play a key role that making a surprise in their structure, these music works typically have a high harmonic surprise, where the music deviates from the listeners’ expectations. Scientists assume that these variations in the structure are recognised as a valuable new information and could elicit a pleasurable reward response in the brain, leading to a positive emotion. The preference of the high surprises in the harmony is an inbuilt setting in our brain, linked to the structure of popular songs. In addition, high surprises followed by a lower-surprise section can contribute to the enjoyment of an unfamiliar piece of music as well. In other words, harmonic surprise can increase the likelihood a song will be a hit.
Another study, back to the hit song science field, focuses on typicality score of pop songs (Askin and Mauskapf, 2017) concludes the similar result. It proposes typicality score as a number that measures how similar the hit was to other songs released during the same time period. Considering an artist’s previous successes, as well the popularity of their record labels and the labels’ marketing budget, chart-topping tracks combined familiarity with novelty. Those songs sounding too much like previous and contemporaneous productions—those that are highly typical—are less likely to succeed. In comparison, songs with a below-average typicality tend to do better on the Hot 100. While bearing some similarity to other popular songs, these songs need to be unique enough to exhibit some degree of optimal differentiation.
Optimally differentiated songs have some combination of acoustic features that are sufficiently dissimilar, but not too different, to that of the other songs released in the previous year. Perhaps the tempo or energy is unlike other songs on the chart around the same time, while other features are relatively similar. The perfect optimally differentiated example needs to stand out from its competition, but not so much as to alienate large groups of potential listeners. These findings offer a new perspective on success in cultural markets by specifying how content organizes product competition and audience consumption behaviour, encouragement to the variation might be the silver lining to the boring sameness of music in the future.
Elberse, A., Eliashberg, J., & Villanueva, J. (2005). Polyphonic HMI: Mixing Music and Math HBS No. 9-506-009, Boston: Harvard Business School Publishing.
Pachet, F. and Roy, P. (2008) ‘Hit Song Science is Not Yet a Science’. Proceedings of ISMIR 2008, pages 355-360, Philadelphia, USA.
Pachet, F.(2011) ‘Hit song science’. In Tzanetakis & Ogihara Tao, editor, Music Data Mining, chapter 10, pages 305–326. Chapman & Hall / CRC Press.
Ni, Y., Santos-Rodriguez, R., Mcvicar, M., De Bie, T.(2011) ‘Hit song science once again a science?’ In: Proceedings of the 4th International Workshop on Machine Learning and Music: Learning from Musical Structure, Sierra Nevada, Spain.
Walter, D. (2017). Data-Driven Business Is Here (And It’s Not What You Think). Available at: http://www.mind-spa.it/2013/11/08/data-driven-business-is-here-and-its-not-what-you-think/ (Accessed 23 Nov. 2017).
Thompson, D. (2017). The Dark Science of Pop Music. The Atlantic. Available at: https://www.theatlantic.com/magazine/archive/2014/12/the-shazam-effect/382237/ (Accessed 23 Nov. 2017).
Huron, D. (2006). ‘Is Music an Evolutionary Adaptation?’. Annals of the New York Academy of Sciences, 930(1), pp.43-61.
Serrà, J., Corral, Á., Boguñá, M., Haro, M. and Arcos, J. (2012). ‘Measuring the Evolution of Contemporary Western Popular Music’. Scientific Reports, 2(1).
Miles, S., Rosen, D. and Grzywacz, N. (2017). ‘A Statistical Analysis of the Relationship between Harmonic Surprise and Preference in Popular Music’. Frontiers in Human Neuroscience, 11.