Michael Krause

PhD in machine learning for music and audio processing. My vision is to create algorithms that can listen to and understand sound. To this end, I build deep learning models and analyze their behavior in depth. Research interests include: audio generation, evaluation of generative models, music information retrieval, deep learning, audio signal processing, self-supervised learning, musical sound event detection, differentiable digital signal processing, soft dynamic time warping.

Publications

• Johannes Zeitler, Michael Krause, and Meinard Müller
  Soft Dynamic Time Warping with Variable Step Weights
  Published in: Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2024.
  Links: PDF
• Simon Schwär, Michael Krause, Michael Fast, Sebastian Rosenzweig, Frank Scherbaum, and Meinard Müller
  A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction
  Published in: Transactions of the International Society for Music Information Retrieval (TISMIR), 7(1): 30–43, 2024.
  Links: PDF, Project page (incl. data)
• Michael Krause, Sebastian Strahl, and Meinard Müller
  Weakly Supervised Multi-Pitch Estimation Using Cross-Version Alignment
  Published in: Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2023.
  Links: PDF, Project page (incl. code)
• Michael Krause, Christof Weiss, and Meinard Müller
  A Cross-Version Approach to Audio Representation Learning for Orchestral Music
  Published in: Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2023.
  Links: PDF, Project page (incl. code, data)
• Johannes Zeitler, Simon Deniffel, Michael Krause, and Meinard Müller
  Stabilizing Training with Soft Dynamic Time Warping: A Case Study for Pitch Class Estimation with Weakly Aligned Targets
  Published in: Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2023.
  Links: PDF
• Michael Krause and Meinard Müller
  Hierarchical Classification for Instrument Activity Detection in Orchestral Music Recordings
  Published in: IEEE/ACM Transactions on Audio, Speech, and Language Processing (TASLP), 2023.
  Links: PDF, Project page (incl. data)
• Christof Weiß, Vlora Arifi-Müller, Michael Krause, Frank Zalkow, Stephanie Klauk, Rainer Kleinertz, and Meinard Müller
  Wagner Ring Dataset: A Complex Opera Scenario for Music Processing and Computational Musicology
  Published in: Transactions of the International Society for Music Information Retrieval (TISMIR), 6(1): 135–149, 2023.
  Links: PDF, Data
• Michael Krause, Christof Weiss and Meinard Müller
  Soft Dynamic Time Warping for Multi-Pitch Estimation and Beyond
  Published in: Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2023.
  Links: PDF, Code
• Michael Krause, and Meinard Müller
  Hierarchical Classification of Singing Activity, Gender, and Type in Complex Music Recordings
  Published in: Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), 2022.
  Links: PDF
• Michael Krause, Meinard Müller, and Christof Weiss
  Towards Leitmotif Activity Detection in Opera Recordings
  Published in: Transactions of the International Society for Music Information Retrieval (TISMIR), 4(1): 127–140, 2021.
  Links: PDF, Project page (incl. data)
• Meinard Müller, Yigitcan Özer, Michael Krause, Thomas Prätzlich, and Jonathan Driedger
  Sync Toolbox: A Python Package for Efficient, Robust, and Accurate Music Synchronization
  Published in: Journal of Open Source Software (JOSS), 6(64): 1–4, 2021.
  Links: Code, PDF
• Michael Krause, Meinard Müller, and Christof Weiss
  Singing Voice Detection in Opera Recordings: A Case Study on Robustness and Generalization
  Published in: Electronics, 10(10): 1–14, 2021.
  Links: PDF
• Michael Krause, Frank Zalkow, Julia Zalkow, Christof Weiss, and Meinard Müller
  Classifying Leitmotifs in Recordings of Operas by Richard Wagner
  Published in: Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), 2020.
  Links: PDF, Poster, Project page (incl. data)
• Kristian Kersting, Christoph Lampert, and Constantin Rothkopf (eds.)
  Wie Maschinen lernen - Künstliche Intelligenz verständlich erklärt
  ISBN: 978-3-658-26762-9,
  Chapters: Maschinelles Lernen (Machine Learning), Bayesregel (Bayes Rule)
  Springer, 2019.
  
• Paul Voigtlaender, Michael Krause, Aljoša Ošep, Jonathon Luiten, Berin Balachandar Gnana Sekar, Andreas Geiger, and Bastian Leibe
  MOTS: Multi-Object Tracking and Segmentation
  Published in: The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019.
  Links: PDF, Project page (incl. data), Code

CV

• 2024–: Research Scientist at Mirelo AI
  • Researching foundational models for audio generation
• 2023–2024: Traveling
• 2019–2023: PhD student at the International AudioLabs Erlangen
  • Music Information Retrieval
  • Supervisor: Prof. Meinard Müller
  • PhD thesis: Activity Detection for Sound Events in Orchestral Music Recordings
• 2017–2019: Postgraduate studies of Computer Science at RWTH Aachen University
  • i.a. machine learning, computer vision, speech recognition and logic
  • Philosophy and ethics of technology
  • Master thesis: Multi-Object Tracking and Segmentation using Instance Embeddings
• 2013–2017: Undergraduate studies of Computer Science at RWTH Aachen University
  • Basic study, algorithmic learning theory, SMT solving
  • Ethics and philosophy of science
  • Bachelor thesis: Sensitivity Analysis over Polymatroids with Applications to Game Theory
• 2015–2016: ERASMUS+ exchange to the University of Edinburgh
  • i.a. quantum computing, algorithmic game theory, machine learning, music informatics and secure programming

Works

• PhD thesis: Activity Detection for Sound Events in Orchestral Music Recordings
  Composers of music can express emotions and communicate with their audience in a multitude of ways. They decide on which voices or instruments to use, arrange notes into melodies, and develop recurring musical patterns. When a composition is performed and turned into sound, their decisions are realized acoustically as sound events. Despite being easily understood by human listeners, teaching a machine to perceive and process such musical sound events can be a challenging task. This thesis studies computational techniques for detecting the activity of sound events in a music recording, i. e., identifying the exact moments in time when a certain event occurs. We focus on orchestral and opera music, which are rarely considered in music processing research and particularly complex due to their high degree of polyphony. In this context, we cover four different types of musical sound events, namely singing, instrumental sounds, different pitches, and leitmotifs (special kinds of musical patterns used for storytelling in opera). To detect the activity of these events within a recording, we design, implement, and evaluate deep learning systems. In addition, we explore a range of techniques including hierarchical classification, differentiable sequence alignments, and representation learning. Beyond evaluating the accuracy of our detection systems, we aim at a deeper understanding of our models with regard to their robustness and sensitivity to confounding effects.
• Master thesis: Multi-Object Tracking and Segmentation using Instance Embeddings
  Simply put, the goal of this thesis is to build and evaluate algorithms that can identify objects in a video by assigning each pixel in the video to the object instance it belongs to, or to mark it as background. This goal draws upon several different research areas of computer vision, the study of algorithms that process visual data like images and videos. Research questions that have previously been considered separately - such as “where in this video is the machine operator’s right hand?” and “which pixels correspond to a hand and which pixels are background?” - now need to be considered jointly.
• Bachelor thesis: Sensitivity Analysis over Polymatroids with Applications to Game Theory
  This thesis deals with certain combinatorial structures and their applications to game theoretic problems. As part of this, we will review several classical results and recent publications on these topics and give illustrating examples. We will give an introduction to matroid and polymatroid theory as well as basic concepts from non-cooperative and cooperative games. Our focus will be on applications of sensitivity results for polymatroid optimization to matroid congestion games and related concepts. In addition, we will demonstrate how the polymatroid structure of certain convex cooperative games can be exploited to recompute core allocations efficiently.
• Music Generation Using Machine Learning
  This report deals with computer generated music. In particular, it discusses two recent approaches for automated melody composition using machine learning methods. Apart from melody, music needs accompaniment (e.g. chords played by different instruments) and must be performed, possibly by a synthesizer. Computer generated music can be used where music is required on- demand (e.g. video games) or as an aid for human composers.
• Quantum Simulation
  This essay deals with a technique called Quantum Simulation, which means simulating one quantum system through another. It begins by motivating the benefits one might obtain from simulating quantum systems and by introducing the main challenges one faces in the process. In section 2, an algorithm that employs quantum simulation to solve linear equations is reviewed and possible applications are discussed.
• Algorithmic Learning Theory - Pattern Languages
  In this report, we have looked at pattern languages. These were originally introduced as a model for inductive inference of formal languages, the question being, if and how a pattern descriptive of a set of strings can be found algorithmically. We studied the results of Dana Angluin, which state that finding descriptive patterns of maximum length is an NP-complete problem. We then described her algorithm for finding descriptive patterns with only one variable. Furthermore, we put the problem into the context of Gold's model of learning in the limit and discussed the algorithm by Lange and Wiehagen. Finally, we described some further results of research into pattern languages.
  Download the slides of the accompanying talk