Reference Audio

Reference audio is a complement to the musical description and lets you provide an audio file as reference for acoustic guidance.

What does it do?: It lets you provide a reference audio clip that acts as a style guide for the music generation model. You are telling the model to make something that acoustically sounds similar to the provided audio but that is still different. You can use this to achieve similar timbre, mixing style or performance characteristics.

What does it not do?: When you add reference audio the acoustic feature information is extracted from the selected file, but specific melody, rhythm, and other structural information is stripped out. Reference audio is not the same as requesting a cover song and can not be used for this purpose. It will not give you a similar song structure, nor can it be used to mimic a specific melody.

You can use reference audio as a complement to the musical description. The musical description provides semantic guidance to the model, whereas the reference audio file can provide acoustic guidance.