Job Description

###

Key Responsibilities

Data acquisition

Develop and run scalable infrastructure for acquiring massive-scale audio (sound and music) and multimodal video-audio datasets Coordinate data transfers from licensing partners and turn heterogeneous sources into training-ready datasets
Annotation and data quality

Obtain detailed annotations for audio and video data (descriptions, musical attributes, audio attributes, …) Use state-of-the-art ML models for data cleaning, processing and filtering Ensure data quality by automated tools and manual evaluation studies Build scalable tools to analyze our datasets (compute statistics, create visualizations, …)
Efficient workflows and collaboration

Optimize and parallelize data processing workflows to handle massive-scale datasets efficiently across both CPUs and GPUs Work directly in the model development loop, updating datasets as training trajectories reveal what we're missing
###

Ideal Candidate Profile

Strong proficiency in Python and experience with various file systems for data-intensive manipulation and analysis Hands-on familiarity with cloud platforms (AWS, GCP, or Azure) and Slurm/HPC environments for distributed data processing Experience with audio and video processing libraries (ffmpeg, …) and an understanding of their performance characteristics Demonstrated ability to optimize and parallelize data workflows across both CPUs and GPUs Knowledge of machine learning techniques for data cleaning and preprocessing
###

Nice to Have

Have built or contributed to large-scale data acquisition systems and understand the operational challenges Have implemented data processing and cleaning pipelines at scale Familiarity with audio and video annotation processes for ML and experience with the specifics of audio data * Have been part of shipping a state-of-the-art model and understand how data decisions impact training outcomes