Member of Technical Staff, Video Generation - Audio

xAI•Palo Alto, CA

11h•Onsite

About The Position

The Omni team at xAI creates magical AI experiences beyond text, enabling understanding and generation of content across various modalities, including image, video, and audio. As a multimodal engineer with a focus on audio, you will drive the model’s audio generation and understanding capabilities, particularly in the context of video and broader media generation. This includes advancing audio for video, music generation, general audio synthesis, speech processing, audio understanding, audio representation / tokenizer, and so on. You will contribute across data curation, modeling, inference serving, and product integration, working on both pretraining and post-training while collaborating with product teams to push the frontiers of model capability and end-to-end user experience.

Requirements

Track record in leading studies that significantly improves the capability and performance of neural networks, whether better data or better modeling.
Experience in data-driven experiment designs and systematic analysis for iterative model debugging.
Experience in developing or working with large-scale distributed machine learning systems.
Ability to do whatever is necessary to deliver the best end-to-end user experience.
Audio-related data processing and model training experience.

Nice To Haves

Experience in graphics engines and rendering techniques is considered an advantage.

Responsibilities

Creating and driving research agendas to advance multimodal audio capabilities, including audio generation for video, audio understanding, audio generation, music generation, and speech generation.
Improving data quality for audio and multimodal datasets, developing advanced annotation, filtering, augmentation, and synthetic data generation techniques, and conducting in-depth data studies.
Designing evaluation frameworks, metrics, and internal benchmarks tailored to audio.
Designing and implementing effective and efficient algorithms for achieving state-of-the-art model performance.