Stems for Wind Instruments

June 13, 2023

As audio researchers and avid music fans, jazz has been one of the most fascinating and eyebrow raising genres for our team to study. Often recorded live and full of improvisations, each performance is distinct and presents a unique opportunity to capture a specific moment in time. AudioShake has been fortunate to work on projects to create immersive mixes of iconic works of jazz musicians like Nina Simone and Tomasz Stańko, as well as master a live album of the legendary Oscar Peterson to isolate and remove bleed between the audience and band mics.

Up until now, our technology has been used by jazz artists and labels like Mack Avenue and Astigmatic to separate clean, high-quality stems from the layers of vocals, piano, bass, and drums that seem to erupt in fits and starts. However, one of the most challenging instruments for source separation tools, and a key component of jazz music, is wind instruments.

Today we’re excited to release stem separation capabilities for wind instruments. Winds have long stumped source separation tools because of their large range of expressions. For example, the playing techniques on a saxophone generate sounds that require audio models with specifically tuned analysis and synthesis modules.

Our AI has spent hours studying different wind instruments and complex compositions to identify the key characteristics that distinguish winds from other stems.

Winds will give us the ability to expand our work in jazz and partner with new and existing AudioShake users on more creative projects. From splitting archival jazz recordings that were made in the days of analog, through to separating multi-genre live tracks today, we’re excited to see what creative possibilities are opened up through wind instrument separation.