Assessing hit song potential is a challenge in the music industry. The question of what song to promote, which song to release first and whether or not it will succeed has always been an issue for stakeholders in the music business. The ability to statistically evaluate hit song potential is a growing field with several studies exploring the topic. In this study we explore the field of hit song science and explore the Swedish hit song profile. We compile hit songs in Sweden between the years 2020 to 2023 and evaluate the distinction between hits and non-hits by using several machine learning methods. The question of assessing lyrics is explored by encoding topic content by using chat-GPT on song lyrics. An accuracy of 70.1% concerning the ability to correctly predict whether a song has been a hit was achieved with the RandomForest algorithm and the feasibility of statistically quantifying hit song potential and its future direction as a research field is discussed. / Att bedöma hitlåtspotential är en utmaning i musikindustrin. Frågan om vilken låt att marknadsföra, vilka låtar som ska släppas först och huruvida de kommer att vara framgångsrika ekonomiskt har alltid varit en utmaning. Förmågan att statistiskt evaluera en låts marknadsmässiga potential är ett växande fält med ett flertal studier. I denna studie utforskar vi fältet “hit song science” och utforskar den svenska hitlåtsprofilen. Vi sammanställer hitlåtar i Sverige mellan åren 2020 till 2023 och utvärderar distinktionen mellan hits och icke-hits med hjälp av maskininlärningsalgoritmer. Vi utforskar förmåga att utröna sångtextsämnen genom att använda chat-GPT på låttexter. En träffsäkerhet på 70.1% uppnåddes via RandomForest-algoritmen och förmågan att statistiskt förutse en låts hitpotential och den framtida forskningsriktningen diskuteras.

Through the analysis of the spectral characteristics of thousands of mastered (or remastered) commercial recordings from a variety of genres over the history of popular music, researchers have studied stylistic trends and spectral conventions. The aim of this study was to further explore, analyse and compare the spectral characteristics of two broad but distinct popular music genres: rock and vocal-based electronic music. The main reason for this choice of genres being that rock generally predominantly is based on (amplified) acoustical elements (e.g. acoustic drums and acoustic/electric bass/guitars) and that electronic music generally predominantly is based on electronic elements (e.g. beats and synthesizers). The stimuli for the study consisted of 24 top-five hit songs from the Billboard charts between 2016-2020, divided by genre. A fast fourier transform approach was used for the computation of target spectrums as well as low level descriptors for the two independent datasets of recordings. Spectral analysis followed with the goal of answering the following research questions: What do the spectral stylistic conventions appear to be in rock versus vocal-based electronic music and what spectral differences/similarities exists between these two distinct popular music genres? The results showed that there were some significant spectral differences between the two genres, especially noticeable in the low end of the frequency spectrum. Other genre-specific spectral trends and overall spectral conventions were found as well.