About this role
Summary: Conduct research in generative audio using diffusion or flow-matching models, focusing on voice generation, speech synthesis, and realtime generation. Build evaluation methodologies and collaborate with engineering to ship new listening experiences.
Join the Computational Music Technologies team, where we build the audio intelligence behind some of Apple’s most distinctive listening experiences. We shipped AutoMix, intelligent DJ-style transitions between tracks, and the Music Understanding Framework, our on-device audio feature extraction system. We’re now building the next generation of generative audio technology, including the foundational models that will underpin a range of new experiences for music lovers and creators across Apple.
Music is our passion, and so is doing the research right: rigorous evaluation, careful attention to artist and listener trust, and a bias toward shipping. Our team sits close to partners in research and engineering across Apple, and we’re given the resources and access to large GPU clusters to enable us to pursue ambitious, foundational work.
Description
As a Machine Learning Researcher on the team, you’ll conduct research in generative audio using diffusion or flow-matching models, with a focus on one or more of the following areas: voice generation, researching vocal and speech synthesis alongside related audio and signal processing techniques; post-training and alignment techniques to improve output quality, controllability, and adherence to human preference; and realtime generation, developing architectures and distillation methods that bring generation latency down to the point where it can run live, including few-step and streaming approaches.
Alongside this, you’ll help build the foundations that everything above depends on: evaluation methodology, benchmarking against the state of the art, and questions of attribution. You’ll run experiments at scale on our own infrastructure, work closely with engineering, product, and design partners to turn research into real listening experiences, and stay connected to the broader research community.
Is this you? If so, we’d love to hear from you.
Minimum qualifications
A PhD in computer science, statistics, applied mathematics or related field
Peer-reviewed publications at venues such as ICASSP, ISMIR, Interspeech, NeurIPS, ICLR, ICML or related
Expertise in generative modelling, machine learning, music information retrieval or speech and audio processing
Industry experience of coding in Python and PyTorch or JAX
Excellent communication and presentation skills
Preferred qualifications
Deep expertise in one of the following areas: audio generation and editing, post-training of diffusion models, voice synthesis
Experience integrating machine learning models into consumer-facing applications
Love of music