SampleMatch
An AI research prototype from Sony CSL Paris that ranks drum samples by how well they fit your track, because auditioning kick_001.wav through kick_947_FINAL.wav is not a creative workflow.
Machine Learning - Sample Match
Product information
Overview
SampleMatch tackles one of music production's great time sinks. You have 900 drum samples, you need one kick and apparently today's creative process will consist of listening to kick_247_final_FINAL.wav. Developed at Sony CSL Paris, SampleMatch uses machine learning to rank drum samples according to how well they fit the music you are already making. Give it a track, let it examine the context and suddenly that enormous sample folder has an opinionated sorting system.
Does This Kick Belong Here
SampleMatch is interested in compatibility rather than simple sonic similarity. Its model was trained using music and drum samples known to belong together, learning to assign a matching score when presented with new combinations. That addresses a surprisingly fuzzy musical question. We have plenty of theory for explaining why a note clashes with a key, but considerably less for explaining why one snare simply feels right while another does not. SampleMatch lets machine learning have a go at the mysterious bit.
Trained on Taste
The research behind SampleMatch uses contrastive learning to capture relationships between musical context and drum sounds. Rather than asking producers to describe every sample with increasingly desperate tags, the system learns from examples of sounds already used together in finished music. The goal is not to declare one objectively correct kick. Taste would riot. Instead, it ranks candidates so producers can reach promising options faster and make the final decision themselves.
Your Sample Folder Has Entered the Chat
SampleMatch is ultimately less about having AI make music and more about making an annoying part of music production less annoying. The producer still writes the track, hears the suggestions and decides whether the machine has excellent taste or should never be trusted with a snare again. That makes it an interesting example of AI as a studio assistant rather than an automatic composer. It does not want your creative job. It would just really like to help with that folder containing 900 kicks.