PhD Research Intern

Join the leader in entertainment innovation and help us design the future. At Dolby, science meets art, and high tech means more than computer code. As a member of the Dolby team, you'll see and hear the results of your work everywhere, from movie theaters to smartphones. We continue to revolutionize how people create, deliver, and enjoy entertainment worldwide. To do that, we need the absolute best talent. We're big enough to give you all the resources you need, and small enough so you can make a real difference and earn recognition for your work. We offer a collegial culture, challenging projects, and excellent compensation and benefits, not to mention a Flex Work approach that is truly flexible to support where, when, and how you do your best work.

Internship: Development and On-Device Deployment of an Autoregressive Model for Real-Time Speech Restoration and Dialog Enhancement

Start: Open

Duration: 3 months (full-time) or 6 months (part-time)

Location: Stockholm

Speech restoration and dialog enhancement are critical components in modern audio experiences, especially in noisy environments or legacy content with degraded quality. Current systems often rely on non-causal, server-side models that require full audio context, limiting their use in real-time or on-device applications. With recent advances in autoregressive language modeling and efficient deep learning deployment, there is growing potential to develop low-latency, high-quality speech enhancement systems that operate in real time and even on consumer devices.

This internship aims to design and implement an autoregressive, large language model (LLM)-based architecture for speech restoration and dialog enhancement. The model will be trained to reconstruct clean speech from degraded inputs by leveraging both acoustic and linguistic context. It will be evaluated against Dolby's current best multimodal system, focusing on perceptual quality, intelligibility, and latency. A key objective is to explore the feasibility of deploying the model on-device through techniques such as model compression, quantization, pruning, and compilation using modern inference frameworks.

• Master's or Ph.D. student in Machine Learning, Signal Processing, Computer Science, or a related field.
• Hands-on experience with PyTorch.
• Interest in model optimization for edge deployment using tools such as Triton Inference Server, ONNX, TensorRT, or similar frameworks. Prior experience in this field is a plus.
• Experience fine-tuning LLMs is a plus.
• Knowledge of audio-language modeling is a plus.