TLDR: Splash Music has launched HummingLM, a groundbreaking AI tool developed in collaboration with AWS, capable of converting vocal hums into complete, studio-quality music tracks. This innovation, powered by AWS Trainium and Amazon SageMaker HyperPod, aims to democratize music creation, making it accessible to millions of creators worldwide.
In a significant leap for generative artificial intelligence in the music industry, Splash Music has officially unveiled HummingLM, an advanced AI-powered model designed to transform simple vocal hums into fully produced music tracks. This innovative tool, developed in close collaboration with Amazon Web Services (AWS), leverages cutting-edge machine learning infrastructure to deliver studio-quality musical compositions.
HummingLM, a proprietary multi-modal generative model, is the result of Splash Music’s participation in the 2024 AWS Generative AI Accelerator program and extensive work with the AWS Generative AI Innovation Center (GenAIIC). The model’s architecture is built around a transformer-based large language model (LLM) integrated with a specialized music encoder upsampler. It utilizes Descript-Audio-Codec (DAC) audio encoding to capture intricate frequency and timbre characteristics from human humming, translating these subtle vocalizations into rich instrumental performances without requiring explicit timbre representation learning.
The primary challenge addressed by Splash Music was scaling music generation to meet the burgeoning demand for unique, instantly created content. The company, which has already facilitated over 600 million streams globally, recognized the need for smarter, faster tools that adapt to evolving user tastes. By partnering with AWS, Splash Music was able to significantly scale its computing power and storage, ensuring HummingLM could handle the complexities of generating high-fidelity audio.
Key to this scaling effort was the deployment of AWS Trainium, Amazon’s custom-designed machine learning accelerator, and Amazon SageMaker HyperPod, a service optimized for distributed training of large machine learning models. These AWS solutions provided the robust infrastructure and performance necessary to accelerate the development lifecycle of Splash Music’s foundation models, enabling continuous adaptation, training, fine-tuning, and deployment of new models.
Also Read:
- Spotify Launches Dedicated AI Lab, Pledges Ethical Development and Artist-Centric Tools
- Fitch Ratings Cautions: AI Music Surge Threatens Artist Royalties and Music IP Investments
This collaboration underscores Splash Music’s mission to make professional music creation truly accessible to a new generation of creators, empowering individuals regardless of their musical skill to bring their ideas to life. The rapid pace of AI advancement in the industry necessitates continuous innovation, and HummingLM represents a new standard in AI-powered music creation, offering intuitive and inspiring experiences for millions.
