Tag: AI

Odd Challenges of Using Deep Learning in Designing a Feedback Delay Network Reverb – Wojciech Kacper Werkowicz & Benjamin Whateley

Join Us For ADC24 - Bristol - 11-13 November 2024
More Info: https://audio.dev/
@audiodevcon​

Odd Challenges of Designing a Feedback Delay Network Reverb With Deep Learning - Wojciech Kacper Werkowicz & Benjamin Whateley - ADC 2023

Past lustrum have seen the rise of interest in optimization of audio effects and synthesizer parameters in use cases including parameter inference from audio input, as well as approaches for Differentiable Digital Signal Processing (such as Magenta's DDSP). However, there are still notable limitations in the area, exemplified well by the problems posed by some fundamental DSP units such as IIR filters - issues of stability, interpretability and differentiability.

In this talk, we will take on all of the above. It will be done so in the context of a research endeavour into modelling room Impulse Responses using Feedback Delay Network (FDNs). Covering a range of approaches, from naive to more advanced, we will take multiple detours to look into machine learning challenges in context of direct applications to DSP, such as approximating common transformations, tackling computational efficiency, taming the explosivity of feedback systems, at last, hopefully, differentiating the undifferentiable.
_

Wojciech Kacper Werkowicz

Programmer, computer musician, improviser from Pruszków, Poland. After being introduced to electronic music by "Ishkur's Guide" in early episode of life, his interest persisted over years. Graduated from Music Computing and Technology BSc program at Goldsmiths in 2023, where he studied under Michael Zbyszynski, Seth Horvitz and Lance Putnam. Currently surveying historical and contemporary digital synthesis methods as a part of his Masters research at Institute of Sonology, The Hague, aiming to critically contextualise synthesis technologies through the lens of sound culture and philosophy. Interested in algorithmic music, machine learning, internet culture. Often enjoys mixing lo-fi technologies with the cutting edge.
_

Benjamin Whateley
_

Streamed & Edited by Digital Medium Ltd: https://online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC23 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Jim Roper
Jonathan Roper
Prashant Mishra

#adc #deeplearning #dsp #audio

Filed under: UncategorizedTagged with: , , , ,

Collaborative Songwriting & Production With Symbolic Generative AI – Sadie Allen & Anirudh Mani – ADC23

https://audio.dev/ -- @audiodevcon​

Collaborative Songwriting and Production With Symbolic Generative AI - Sadie Allen & Anirudh Mani - ADC23

Generative AI has experienced remarkable advancements in various domains, including audio and music. However, despite these breakthroughs, we have yet to reach a stage where musicians can seamlessly incorporate generative AI into their creative processes. In this talk, we will delve into the techniques, proposals, and ongoing work that can facilitate collaborative songwriting and production with machine learning.

During the session, we will explore several key topics:
• Overview of existing tools and models - we will discuss the motivations behind symbolic generation versus raw audio for music production applications. Furthermore, we will highlight the contrasting approaches and techniques that aim to augment the creative process rather than replace it entirely.
• Utilization of AI-generated MIDI as a songwriting tool - this will involve examining different ML architectures for conditional MIDI generation, as well as employing reinforcement learning (RL) to generate MIDI sequences.
• Examples showcasing how speakers and other musicians currently utilize AI-generated MIDI as part of their songwriting/production process.

Attendees will gain insights into cutting-edge techniques and research, paving the way for a future where generative AI becomes an integral part of the creative process for musicians.

Link to Slides: https://drive.google.com/file/d/15qYW-SbgmodMZ_wiMKKvH8pXmrDCZQpY/view?usp=sharing
_

Sadie Allen
_

Anirudh Mani

I build creative AI tools for artists.I am the co-founder of Lemonaide Music. https://www.lemonaide.ai/ https://www.linkedin.com/in/anirudh-mani-1796934b/ https://twitter.com/anirudh3
_

Streamed & Edited by Digital Medium Ltd: https://online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC23 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Jim Roper
Jonathan Roper
Prashant Mishra

#adc #ai #dsp #audio #generativeai

Filed under: UncategorizedTagged with: , , , ,

Motion and Music Modeling in Hindustani Classical Music – Tejaswinee Kelkar – ADCx India 2024

Join Us For ADC24 - Bristol - 11-13 November 2024
More Info: https://audio.dev/
@audiodevcon​

Motion and Music Modeling in Hindustani Classical Music - Tejaswinee Kelkar - ADCx India 2024

My talk will summarize of computational generative approaches in North Indian classical music (NICM). NICM presents a unique problem where non-quantization of notes, and the predominant characteristic use of pitch contours to express sonic differentiation means that quantized modeling of, for example, sheet based music goes only so far in being able to shape generative Hindustani music. I will present these approaches of notation based, and character based RNNs for generating Hindustani improvisation.

Generative musical AI in NICM is not really described as a task. However, pre-trained generative music models are modeled after common practise period based western music, and are definitely unsuitable to generate anything in this vocabulary. Sample based generative AI for NICM has, as of this abstract not been a field with separate exploration. Musical AI in NICM is mostly explored form the point of view of modeling raga and raga recognition tasks.

In my previous work, I have addressed how phrase generation models and contour models are perceptually important for tasks such as this. I will present an overview of the state of knowledge in the intersection of these fields and the SOTA of generative techniques in NICM.

Link to Slides: https://data.audio.dev/talks/ADCxIndia/2024/rnns-and-hindustani-music.pdf
_

Edited by Digital Medium Ltd - online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC24 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Prashant Mishra

#adc #ai #audio #hindustaniclassicalmusic

Filed under: UncategorizedTagged with: , , , ,

Developing an AI-Powered Karaoke Experience – Thomas Hézard & Clément Tabary – ADC23

https://audio.dev/ -- @audiodevcon​

Developing an AI-Powered Karaoke Experience - Thomas Hézard & Clément Tabary - ADC23

Karaoke has been of popular interest for many years, from the first karaoke bars in the 1970s to the karaoke video games of today, and the recent progress in deep learning technologies has opened up new horizons. Audio source separation and voice transcription algorithms now give the opportunity to create a complete karaoke song, with instrumental track and synchronised lyrics, from any mixed music track. Real-time stems remixing, pitch and tempo control, and singing quality assessment are other useful audio features to go beyond the traditional karaoke experience. In this talk we will discuss the challenges we had to tackle to provide our users with a fully automatic and integrated karaoke system adapted for both mobile and web platforms.
_

Thomas Hézard

Thomas leads the Audio Research & Development team at MWM, working with his team on innovative signal processing algorithms and their optimised implementation on various platforms. Before joining the MWM adventure, Thomas completed a PhD on voice analysis-synthesis at IRCAM in Paris. Fascinated by every aspect of sound and music, both artistic and scientific, Thomas is also a musician, a sound engineer, a passionate teacher, and an amateur photographer.
_

Clément Tabary

Clément is a deep-learning research engineer at MWM. He applies ML algorithms to a wide range of multimedia fields, from music information retrieval to image generation. He's currently working on audio source separation, music transcription, and automatic DJing.
_

Streamed & Edited by Digital Medium Ltd: https://online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC23 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Jim Roper
Jonathan Roper
Prashant Mishra

#adc #audiodev #ai #karaoke

Filed under: UncategorizedTagged with: , , ,

Virtual Studio Production Tools With AI Driven Personalized Spatial Audio for Immersive Mixing

Join Us For ADC24 - Bristol - 11-13 November 2024
More Info: https://audio.dev/
@audiodevcon​

Virtual Studio Production Tools With AI Driven Personalized Spatial Audio for Immersive Mixing - Dr. Kaushik Sunder & Krishnan Subramanian - ADCx India 2024

In recent years, Spatial audio formats such as Dolby Atmos, Sony 360 Reality, Auro 3D are on the rise. As a result of this, there is also an increasing need for having multi channel speaker setups and associated gear in the studio to produce, mix, and master music in such formats. These systems are extremely expensive, occupy space, time consuming to set up, and therefore a massive barrier to entry for most mixing engineers. In this talk, we will present some of the latest innovations in enabling an ecosystem of Virtual Studio Production with AI driven personalized spatial audio. We explore the need and integration of personalized HRTFs, Room acoustics modeling, and personalized headphone equalization for such virtual production tools. We will also present our experience leveraging JUCE for building spatial audio plugins, particularly as it pertains to virtualizing real world acoustic environments. By sharing our insights, this talk aims to provide valuable information to developers interested in building spatial audio plugins that bring down barriers of cost, accessibility, making “immersive for all” a reality for creative professionals.

Link to Slides: https://data.audio.dev/talks/ADCxIndia/2024/ai-driven-personalized-spatial-audio-for-immersive-mixing.pdf
_

Edited by Digital Medium Ltd - online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC24 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Prashant Mishra

#adc #ai #audio #virtualstudio

Filed under: UncategorizedTagged with: , , , ,

AI Generated Voices: Towards Emotive Speech Synthesis – Vibhor Saran – ADCx India 2024

Join Us For ADC24 - Bristol - 11-13 November 2024
More Info: https://audio.dev/
@audiodevcon​

AI Generated Voices: Towards Emotive Speech Synthesis - Vibhor Saran - ADCx India 2024

Traditionally, machine generated voices were synthesised by joining the phonemes of any language, which made these voices robotic in nature. With the availability of more data and advent of deep learning, these AI voices started becoming more human and engaging. The next step is to make these AI generated voices more emotive so that it can laugh, be sad or even cry just like how expressive human speech is. In this talk, we touch base upon deep learning approaches to make synthetic voices more emotive. Specifically, we will focus on how to manipulate the Mel Spectrogram of the speech to make it engaging, removing the dependency of large quantums of data.

Link to Slides: https://data.audio.dev/talks/ADCxIndia/2024/towards-emotive-speech-synthesis.pdf
_

Edited by Digital Medium Ltd - online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC24 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Prashant Mishra

#adc #ai #dsp #audio #speechsynthesis

Filed under: UncategorizedTagged with: , , ,

Building AI Music Tools: An Engineer’s Guide to Prototyping – Jamie Pond – ADC23

https://audio.dev/ -- @audiodevcon​

Building AI Music Tools for the 99%: An Engineer’s Guide to Prototyping - Jamie Pond - ADC23

How to go from idea, to lo-if prototype, to validation, to hi-fi prototype to production.
Exploring the method we used to develop and ship 3 large appeal consumer audio apps this year, to millions of users.

Link to Slides: https://data.audio.dev/talks/2023/an-engineers-guide-to-prototyping/slides.pdf
_

Jamie Pond
_

Streamed & Edited by Digital Medium Ltd: https://online.digital-medium.co.uk
_

Organized and produced by JUCE: https://juce.com/
_

Special thanks to the ADC23 Team:

Sophie Carus
Derek Heimlich
Andrew Kirk
Bobby Lombardi
Tom Poole
Ralph Richbourg
Jim Roper
Jonathan Roper
Prashant Mishra

#adc #audiodev #ai #audio

Filed under: UncategorizedTagged with: , , , , , ,