Key results
The company
Splash Music
splashmusic.comAI music creation platform and interactive virtual performance space for gaming.
Result highlights
- 730M+ total streams for AI-generated songs
The story
An Australian music and gaming startup enables users to create full songs by humming a tune, rather than relying on standard text-to-audio prompts.
A lean engineering team struggled with fragmented infrastructure that slowed down experimentation cycles for their generative audio models. They needed to scale training and reduce inference costs without dedicating excessive resources to managing compute clusters.
The company migrated its HummingLM model to AWS Trainium chips for training and deployed inference on EC2 Inf2 instances to optimize performance. SageMaker HyperPod orchestrates distributed training across nodes, while Amazon S3 handles 2 PB of migrated data. The user-facing application runs on AWS Amplify to connect frontend experiences with the backend audio models.
Scope & timeline
- 30,000+ new songs created via HummingLM
Quotes
“We wanted the robustness of AWS compute instances, but also a single solution for hosting, APIs, deployment, and security. AWS gave us that. We wanted to spend less time managing infrastructure and more time building features.”