Soumya Sai Vanka

Research and Development Intern, Yamaha Corporation

• Working on mixing workflow personalisation for mixing consoles, as part of the R&D Division project “Analyzing Individuality in the Workflows of Mixing Music.”
• Researching the structural analysis and modelling of mixing workflows to support more efficient and creative music production.
• Analysing how engineers apply audio effects and construct sound imagery through effect chains and track selection during mixdown.
• Applying machine learning to model the individuality and consistency of each engineer’s practice, informing customised, personalised mixing console systems.
• Collaborators: Yu Takahashi, Hayato Yamakawa, Kazunobu Kondo (MixerAI team).

PhD Researcher, Queen Mary University of London

• Designed, trained and evaluated deep learning models for controllable music mixing style transfer.
• Built large-scale synthetic audio datasets for training and benchmarking ML systems.
• Led user-centred ML research with professional mixing engineers to define product requirements and validate usability.
• Designed interaction workflows for ML-based multitrack mixing systems integrated into DAWs.
• Conducted large-scale UX studies and listening tests to evaluate model behaviour in real creative workflows.
• Collaborated with industry partners at Steinberg Media Technologies on future product-facing research.
• Mentors and collaborators: George Fazekas (QMUL), Jean-Baptiste Rolland (SMTG), Yvan Grabit (SMTG).

Intern, Sony Research

• Designed ML representations for music mixing effect chains using graph neural networks and contrastive learning.
• Built data pipelines for large-scale multitrack scraping and synthetic mix generation.
• Conducted computational research on representation learning for audio post-production workflows.
• Collaborators: Marco Martinez (Supervisor), Sunghoo Lee (Seoul National University), Junghyun Koo, Wei-Hsiang Liao, Yuki Mitsufuji.

Trainee Intern, Steinberg Media Technologies GmbH

• Developed deep learning systems for context-aware music mixing using PyTorch on Kubernetes-based infrastructure.
• Implemented differentiable audio effects and VST plugins for real-time ML-based processing.
• Led dataset design for multitrack mixing research used across academic and industry projects.
• Collaborators: Jean-Baptiste Rolland, Lennart Hannink, Yvan Grabit.

PhD in AI and Music, Queen Mary University of London

Modules: Deep Learning for Music, Machine Learning, Sound Recording and Production, Music Informatics. Fully funded by UKRI AIM-CDT and Steinberg Media Technologies GmbH.

MSc Physics, Pondicherry University

Modules: Non-Linear Dynamics, Electronics, Mathematical Physics. Funded by a Merit Scholarship at PU and the DST-Inspire Scholarship.

BSc Physics (Hons), Sri Sathya Sai Institute of Higher Learning

Modules: Electronics, Mathematical Physics, Set Theory. Fully funded by SSSIHL and the DST-Inspire Scholarship.

Teaching Fellow, Queen Mary University of London

• Supervised 17 undergraduate final-year projects in machine learning and software development across two years.
• Guided students on research design, ML experimentation, and academic/technical communication.
• Themes included ML-based audio systems, computer vision apps, game platforms, and accessibility tools.
• Supported taught modules as a teaching assistant: Professional and Research Practice (UG), Ethics, Regulation and Law in Advanced Digital Information Processing (PG), User Experience Design (UG/PG), Research Methods and Responsible Research (PG).
• Notable project: Personalised Audio Deepfake Detection (Egor Vert), a two-stream WavLM and AASIST detector with per-speaker personalisation via FiLM conditioning, evaluated on the ASVspoof 2019 LA dataset.