Integral scherzo: how machines compose and sing
summary
Overview of modern generative-music technologies for analyzing, composing and singing music with machine learning, including SOTA audio demos and curated example comparisons. Covers practical challenges and surprising findings from R&D at SberDevices and presents their released AI-music technology, with demonstrations and applications rather than heavy math or algorithms.
AI-written summary · about
chapters (25)
- 0:00Introduction
- 1:46Human or neural network test
- 2:53Music and AI
- 5:00AI milestones in music
- 8:37Illiac Suite and musical dice
- 10:03Rudolf Zaripov's algorithmic music
- 11:10Data types in music tasks
- 12:35Emotion-based music classification
- 14:59Audio digitization and spectrograms
- 17:53Convolutional networks for music
- 19:39Audio synthesis and resynthesis
- 23:31Current capabilities
- 24:12Stable Audio and text prompts
- 27:05Music boxes and player pianos
- 29:42Generative models for music
- 31:40MIDI standard and symbolic generation
- 33:25REMI format for sheet music
- 35:11Symformer model capabilities
- 37:15Neural singing and voice conversion
- 39:47Voice synthesis without source audio
- 41:02Song generation from lyrics
- 43:38Real-world applications
- 44:52Real-time and improvisation
- 46:05Future outlook and examples
- 47:13Q&A
related talks
01h 04mKeynote 3. Interactive generative models for music
51mThe Merger of Machine Learning, Music and Programming
44mWith AI, Does Music Have the Blues?
43mMaking machines that make music
23mClosing Keynote: If AI Can Create Music, What Does That Mean...
44mCoding to the Beat: When AI Joins the Band
41mMaking Machines that Make Music
30mDeveloping AI that enhances human creativity