AIVA: Introducing Lyra

In this post, three state-of-the-art models developed in-house are introduced:

• OmniCodec, a neural audio codec

• Lyra, an end-to-end, highly personalised instrumental-only foundation generative model

• ORAQL, a music quality assessment model that closely aligns with the subjective preferences of human listeners.

In addition, Influence, a post-training variant on Lyra, unlocks the popular use case of uploading a reference audio file and generating a completely original composition that shares core musical characteristics with the reference.

More information and audio examples here.