Staff Research Scientist (Diffusion)

Fabrik Talent; United States; Remote; GBP 80,000 - 120,000; Full time

Job summary

A generative AI company is seeking a Staff Research Scientist specializing in audio generation. This role involves designing and training diffusion transformer models for high-fidelity audio. The ideal candidate will have a strong research background in generative models, significant experience in pre-training large models, and a passion for pushing the boundaries of audio technology. The position offers exciting opportunities in frontier research with serious funding and autonomy.

Qualifications

• Contributions to notable audio diffusion projects are a strong plus.
• Strong research background required in generative models.
• Experience pre-training models from scratch is crucial.
• Deep understanding of sequence or signal modeling is needed.

Responsibilities

• Design and train large-scale diffusion transformer models for audio.
• Push the frontier of controllable, high-quality audio generation.
• Own the pre-training strategy including data and objectives.
• Translate research ideas into models for real products.
• Collaborate with research, infra, and product teams.

Skills

Contributions to prominent diffusion projects

Research background in generative models

Experience pre-training large models

Deep understanding of sequence or signal modelling

Comfortable debating papers and trade-offs

Opinionated and curious mindset

Job description

Staff Research Scientist (Audio Generation)

Remote | Text-to-Audio | Ex-DeepMind Team

We’re working with a well-funded generative AI company building an interactive audio platform for both consumers and commercial partners. The founding research team were lead contributors on the most prominant audio generation projects at Google DeepMind. They have signed partnerships in place.

This role sits right at the centre of their core research: pre-training diffusion-based transformer models for high-fidelity audio generation.

What you’ll work on

• Designing and training large-scale diffusion transformer models for audio
• Pushing the frontier of controllable, high-quality audio generation
• Owning pre-training strategy: data, objectives, architectures, and scaling
• Translating research ideas into models that power real interactive products
• Collaborating closely with research, infra, and product to shape the platform

What they’re looking for

• Contribution to some of the most prominent diffusion projects (audio is a strong plus)
• Strong research background in generative models (diffusion, transformers, or both)
• Experience pre-training large models from scratch
• Deep understanding of sequence or signal modelling
• Comfortable debating papers, assumptions, and trade-offs at depth
• Opinionated, curious, and excited about shipping research into the real world

Why this is compelling

• Frontier research with clear product pull
• Audio generation is the core, not a side project
• Serious funding, elite peers, and real autonomy
• Opportunity to define how next-gen audio models are trained and used