Estimating Tonal Tension from Audio Content
|Title||Estimating Tonal Tension from Audio Content|
|Year of Publication||2009|
Detecting emotions from music based on audio content is a relatively recent problem. Most of the approaches tend to categorize the whole musical piece and only a few focus on the time evolution of the emotion.
In this study, we investigate one particular emotional aspect: tension. Moreover, we narrow our work to model a particular component of it: the tonal tension.
We extract tonal information from the audio content in the form of Harmonic Pitch Class Profiles (HPCPs). Then, from this sequence of HPCP vectors, we compute the distances one by one based on Harte's distance (projection of the 12 tones in a 6-D space and euclidean distance). We also extract informations from this curve computing the mean, variance and other statistical measures.
We find correlations comparing the tonal tension model from Lerdahl and the tension curves we extract from the audio content. We also find correlations between statistical descriptions of the tension curves and tension ratings from listeners.
This technique offers promising results and a simplistic model of tonal tension. Future works will consist in enhancing this model to get a more accurate representation of the tonal tension.