Research Scientist Intern, Multimodal LLM and Speech-to-Speech Translation (PhD)

Meta•Redmond, WA

9d•$7,650 - $12,667

About The Position

Meta Reality Labs Research seeks Research Scientist Interns to advance multimodal large language models (MLLMs) and speech-to-speech translation (S2ST) for next-generation wearable and AR/VR experiences. The Reality Labs Research audio team develops technologies intended to help people understand their surroundings, communicate in challenging listening environments, and connect across languages. Research areas include models integrating audio, speech, vision, and language for multimodal understanding and reasoning; conversational speech models; and streaming speech translation that preserves meaning, speaker characteristics, and expressive prosody. Interns investigate research questions in one or more of these areas, develop and evaluate novel models, and work with researchers and engineers to demonstrate their potential in real-world applications. Internships last 12 to 24 weeks and have various start dates throughout the year.

Requirements

Currently has, or is in the process of obtaining, a PhD in Computer Science, Artificial Intelligence, Machine Learning, Electrical Engineering, Signal Processing, Computational Linguistics, or a related STEM field.
3+ years of programming experience in Python or a similar language.
3+ years of experience with machine learning frameworks such as PyTorch or TensorFlow.
2+ years of experience developing computational models in one or more of speech processing, natural language processing, multimodal learning, or audio-visual learning.
Must obtain work authorization in the country of employment at the time of hire and maintain ongoing work authorization during employment.

Nice To Haves

Research experience in multimodal large language models, including audio-language or audio-visual-language understanding, alignment, and reasoning; speech-to-speech translation, simultaneous translation, multilingual speech modeling, or streaming speech generation; or speech foundation models, neural audio codecs, speech tokenization, or expressive speech synthesis.
Experience pretraining or fine-tuning large models using distributed training and large-scale datasets.
Experience evaluating multimodal reasoning or speech translation, including translation accuracy, speech naturalness, speaker similarity, and streaming latency.
Research contributions through publications at venues such as NeurIPS, ICML, ICLR, ACL, EMNLP, CVPR, ICCV, ECCV, ICASSP, or Interspeech.
Software engineering experience through internships, research projects, or contributions to open-source repositories.
Experience collaborating and communicating across disciplines.
Intent to return to a degree program after completing the internship/co-op.

Responsibilities

Conduct original research in multimodal large language models, speech foundation models, or speech-to-speech translation.
Develop models that jointly understand and reason over audio, speech, visual, and textual inputs, or generate natural speech across languages.
Investigate methods for model pretraining, multimodal alignment, instruction tuning, or streaming inference.
Design and execute experiments, establish baselines, and evaluate model quality, robustness, latency, and computational efficiency.
Collaborate with researchers and engineers across speech, audio, computer vision, and language modeling.
Communicate research ideas, progress, and results through presentations, technical reports, and research publications.