ELEC-E5632 - Audio Technology Seminar V D, Lecture, 27.2.2023-22.5.2023

Topic outline

• General
  • 

    Basic info Forum

    In spring 2023, the theme of the Audio Technology Seminar will be "Deep Learning for Audio and Music Processing". The seminar will be suitable for a wide range of Acoustics and Audio Technology students, including MSc students who have not yet taken machine learning courses, but are willing to learn, and doctoral researchers who are familiar with and using deep learning methods. The seminar will run on periods IV and V, starting on Feb. 27, 2023, and ending in late May 2023. Ricardo Falcon and several other experts, who are applying deep learning for music and audio in their work, will provide introductory lectures. The seminar will be organized by Prof. Vesa Välimäki, Ricardo Falcon, Eloi Moliner, and Alec Wright. 
    
    The seminar started on Monday, February 27, 2023, at 14.15 in hall U8 (Otakaari 1, Espoo). All lectures of this seminar will take place in the same hall, U8.
     
    
  • Not available unless any of:

    • You are a(n) Student
    • You are a(n) Teacher

    Seminar Schedule 2023

    • L0 (Feb 27): Opening of the seminar, selection of the seminar paper topics (Vesa Välimäki, Eloi Moliner, Alec Wright)
    • L1 (Mar 6): Lauri Juvela, Neural Synthesis of Speech and Musical Instruments
    • L2 (Mar 13): Ricardo Falcon, Introduction to Data Science for Audio and Music Processing and Piano Music Modeling with the MAESTRO Dataset
    • L3 (Mar 20): Alec Wright, Deep Learning for Audio Effects Processing.
      Vesa Välimäki, How to Write a Good Paper.
    • L4 (Mar 27): Eloi Moliner, Generative Models for Audio Enhancement.
      Vesa Välimäki, How to Make a Good Presentation.
    • L5 (Apr 3): Eero-Pekka Damskägg (Neural DSP Technologies), Neural DSP Technologies
      (No lecture on April 10, which is Easter Monday.) 
    • L6 (Apr 17): Giorgio Fabbro (Sony, Germany), Pushing the Boundaries of Audio Production: Deep Learning for Automatic Music Mixing and Audio Source Separation
    • L7 (Apr 24): Tom Lidy (Utopia Music, Austria), Real-world Deep Learning Applications for Music Tagging & Recommendation

    Student Presentations (May)

    • P1 (May 8): Presentations, part 1
    • P2 (May 15): Presentations, part 2
    • P3 (May 22): Presentations, part 3

    
    

  • Grading

    Grades are from 0-5. Criteria for grading include:

    • Attendance
    • Active Participation
    • Punctuality with Submissions
    • Presentation Quality
    • Report Quality

    Seminar paper template

    The expected length of the seminar paper for MSc students is 10-15 pages, and for doctoral students 15-20 pages.
    

    • Please use the template for the seminar paper.

    Submission schedule

    • Fri 17.3. (16.00) Deadline for the Table of contents + some references
      
    • Fri 31.3. (16.00) Deadline for 1st draft (some text + refs)
    • Fri 14.4. (16.00) Deadline for 2nd draft (more text + figures)
    •  ̶M̶o̶n̶ ̶2̶4̶.̶4̶.̶ Wed 26.4 (16.00) Full paper submission (NEW DEADLINE!)
    • End of April. Peer review
    • Fri 28.4. Thu 4.5. (16.00) Presentation slides submission (NEW DEADLINE) 
    • Fri 5.5. (16.00) Deadline for the Learning Diary based on the lectures (min 5 lectures)
    • May 2023. Student presentations. The final paper should be submitted before the seminar presentation.

    
    

  • • Gloria: Differentiable digital signal processing for sound synthesis (DDSP)
    • Tantep: Differentiable FIR and IIR filters

    Seminar paper/presentation topics

    1. Deep learning for audio effects processing (May 8, instructors: Alec & Vesa)
    

    • These student presentations were postponed and distributed for the next two Mondays, see below.

    2. Deep learning for audio generation and enhancement (May 15, instructors: Eloi & Vesa)
    

    • Wu: Diffusion models for audio (and speech) enhancement
      
    • Yifu: Voice conversion using deep learning
    • Ossi: Audio inpainting
      
    • Anja: Dereverberation
      
    • Samu: Denoising
    • Antoni: Blackbox modeling using convolutional neural networks (CNNs)
    • Håvard: Blackbox modeling using recurrent neural networks (RNNs)

    3. Deep learning for musical signal processing (May 22, instructor: Ricardo)
    

    • Ido: Symbolic music generation with transformers
    • Jackie: Learning neural acoustic fields (NERF)
    • Markus: Musical beat detection
    • Meri: Automated Audio Captioning
    • Gloria: Differentiable digital signal processing for sound synthesis (DDSP)
    • Tantep: Differentiable FIR and IIR filters
    • Teodors: Neural audio compression
    • Jon: Generative adversarial networks (GANs) for audio generation
    • Bryn: Audio super-resolution (bandwidth extension)