Research Scientist - Generative Audio

What You'll Do

Conduct groundbreaking research in generative audio using diffusion or flow matching models, with a focus on one or more of the following areas:
Vocal Synthesis — Research in vocal and speech synthesis, along with related areas such as ML-based audio processing and signal processing.
Post-Training — Research in post-training techniques for music generation, including preference alignment methods (such as DPO, RLHF, or KTO), reward model design and training, and reinforcement learning to improve output quality, controllability, and human preference adherence.
Editing — Research in iterative music generation and audio editing, including capabilities such as stem replacement, instrumentation change, mood changes (while preserving content), tempo changes, and structure changes.
You will also:
Run large-scale experiments with access to Spotify's extensive infrastructure and an audience of more than 700 million monthly active users.
Create practical applications that harness generative technologies and push the boundaries of what's possible in listening experiences.
Collaborate as part of a cross-functional team—working closely with scientists, engineers, product managers, designers, user researchers, and analysts—to craft innovative solutions to complex challenges.
Have a direct impact on Spotify's products, tools, and services, working on projects that influence the entire organization.
Engage with the broader research community by publishing your findings, delivering talks, and attending top conferences.

Who You Are

You have a Ph.D. in Computer Science, Mathematics, Engineering, or a related field. Previous industry experience is helpful.
You have experience in one or more of the following fields: generative modeling, machine learning, music information retrieval, speech processing, audio processing, signal processing, probabilistic modeling, computer vision, or related areas.
You have deep expertise in at least one of the focus areas above—whether that's vocal/speech synthesis, post-training alignment techniques (e.g., PPO, GRPO, DPO), or audio-to-audio generation and text-guided music editing.
You have publications at leading conferences such as ICASSP, ISMIR, INTERSPEECH, ICLR, AAAI, IJCAI, NeurIPS, ICML, CVPR, ECCV, ICCV, or related venues.
You have strong coding skills in Python, PyTorch, and NumPy.
You are a creative problem solver who is passionate about building outstanding products that add real value to millions of people.
You are enthusiastic about turning research ideas into products operating at scale.
You can explain complex topics in simple terms, and you enjoy building strong relationships with colleagues and stakeholders.

Where You'll Be

We offer you the flexibility to work where you work best! For this role, you can be within the North Americas region as long as we have a work location.
This team operates within the Eastern Standard time zone for collaboration.
Core working hours are CET 3pm-6pm / EST 9am-12pm.