In the thesis “Philipp Kramer: Relevanz cepstraler Merkmale für Vorhersagen im Arousal-Valence Modell auf Musiksignaldaten” (Relevance of cepstral features for predictions in arousal-valence model for music signals, PDF in German), several groups of features were analysed for the regression-based prediction of arousal and valence: cepstral, energy, timbral, harmonic/melodic, and temporal/rhythmic. Some parameters for the feature extraction were optimised, and it was shown that features from the cepstral domain belonged to the best models.
Prof. Dr. Gerald Langner from TU Darmstadt will give a talk “Der neuronale Code von Tonhöhe, Klang und Harmonie” (The neural code of pitch, sound, and harmony) at the Institute of Music and Music Science, TU Dortmund on 6.7.2016, 14-16 PM, Emil-Figge-Str. 50, R 4.313.
Weihs, C., Jannach, D., Vatolkin, I., Rudolph, G. (Eds.): Music data analysis: foundations and applications.
This book edited and co-authored by SIGMA members and several partners provides a comprehensive overview of music data analysis, from introductory material to advanced concepts. It covers various applications including transcription and segmentation as well as chord and harmony, instrument and tempo recognition. It also discusses the implementation aspects of music data analysis such as architecture, user interface and hardware. It is ideal for use in university classes with an interest in music data analysis. It also could be used in computer science and statistics as well as musicology.
In winter term 2016/2017, two proseminars will take place at Chair of Algorithm Engineering, TU Dortmund:
Jannach, D., Kamehkhosch, I., Bonnin, G.: Biases in Automated Music Playlist Generation: A Comparison of Next-Track Recommending Techniques, User Modeling, Adaptation and Personalization (UMAP 2016), Halifax, CA, 2016
In this work, the results of a multi-metric comparison of different academic approaches and a commercial playlisting service (of The Echo Nest) are reported. The results show that all tested techniques generate playlists with certain biases, e.g., towards very popular tracks, and often create playlists continuations that are quite different from those that are created by real users.
Florian Treinat: Verwendung von Lyrics zur Generierung von Musik-Playlisten (Application of lyrics for generation of music playlists) (supervisors: Dietmar Jannach, Iman Kamehkhosh, e-Services Research Group, TU Dortmund)
The goal of this work was to improve the quality of music recommendations with the help of lyrics. The proposed approaches are based on (a) the textual similarity and (b) the conveyed sentiment of lyrics. The results show that lyrics-based techniques are more efficient when the seed tracks (e.g., the tracks from the recent listening history of the user) are thematically related or sentimentally homogeneous.
Mike Gösker: #nowplaying: Analyse musikbezogener Twitterdaten (Analysis of music-related Twitter data) (supervisors: Dietmar Jannach, Lukas Lerche, e-Services Research Group, TU Dortmund)
In his master’s thesis Mike Gösker implemented and evaluated a set of techniques to generate music track recommendations based on user posts and profiles from the social networking service Twitter. The recommendation strategies exploit temporal characteristics of the social media posts and are compared with baseline techniques that use popularity and neighborhood information.
The goal of the thesis (PDF in German) was to develop a system for the automatic generation of drum accompaniment to improvised Jazz solos. This is solved by means of an evolutionary algorithm. A list of rules is defined for the evaluation of percussive patterns with regard to music and Jazz theory. Furthermore, the thesis provides an overview of related research works.
This thesis addressed the problem of automatically rearranging any given music piece based on user-defined constraints. Rearranged music pieces are generated by playing back the original piece and jumping from one position in the original to another at specific times. The related optimisation problem is defined as the reduction of costs in a bixel path, i.e. jumps between short music segments starting and ending at identified beat events.
In winter term 2015/2016, a proseminar “Actual challenges in music data analysis” (website in German) will take place at Chair of Algorithm Engineering, TU Dortmund. The topics for student talks and works should represent various MIR research areas and are selected from the proceedings of ISMIR 2014.