Topic outline

  • Deep Learning of Audio

    Artificial intelligence (AI) methods have been the most recent trend for artists to experiment with advanced technologies. The growing use of AI in music to support musical creativity enables new possibilities of utilising new musical instruments and interfaces. In Deep Learning of Audio course, we will introduce students to the state of the art in deep learning models and AI methods for sound and music generation. The course will provide an overview of recent AI implementations such as, Google Magenta's (AI Duet, NSynth, GANSynth, DDSP) GANSpaceSynth and SampleRNN. There will be hands-on exercises on each course topic. We will provide code templates that integrate the functionality from open source deep learning audio projects into Pure Data programming environment. We will also provide detailed setup instructions and automated scripts to make installation of the required tools as easy as possible (for Pure Data, Python, Conda, Magenta, PyExt). Students will further explore a particular model and incorporate it into their own project work.

    During the 3 weeks course period, students will have 24/7 access to the Azure cloud computing VMs in Aalto University and will be able to train their deep leaning models for the course exercises as well as for their projects. The current module of installations and course exercises regarding the content generation only work in Linux. There are already some laptop computers with Ubuntu 20.04 reserved for the students of this course in TakeOut / Väre. You should contact in person and get one before the course starts on Tuesday 31.01, if you do not have any linux computers. 

    Students will further explore a particular model and incorporate it into their own project work. Deep Learning with Audio is a project-based course, we dedicate half of the contact hours for project work, the lecturer and the teaching assistants will support students by giving sufficient guidance, feedback and tutoring. At the end of the course, students will submit and present their music composition projects.

    In Deep Learning of Audio course, students will be able to:

    1. Develop a depth of understanding of the recent audio domain deep learning models, artificial intelligence (AI) methods and network architectures.
    2. Build AI applications for their own purposes.
    3. Prepare data sets and train deep learning models using cluster network in Aalto University.
    4. Develop a depth of understanding of the differences in input, computational cost and sonic characteristics between the different models

     

    Locations : Väre L101 / L102 Ryhmäopetus 

    Credits : 3 ECTS ≈ 81h

    Contact hours :  36h 

    Assessment Methods and Criteria:

    The course consists of lectures, exercises, reading materials, tutoring individual or group works. Contact teaching hours are allocated for project work in the classroom and students receive sufficient guidance, feedback and tutoring.


    Active participation in the course – interaction, QA (10%)

    Course assignments (10%)

    Project Interim - Concept presentation (1 page)(10%) 

    Project Final Presentation (15%)

    Project Final Delivery of the Music Composition Project including the video demonstration (40%)

    Project Final Concept paper / learning diary (1-3 page / ~750 words) (15%)