Latest models, higher quality stems
Today AudioShake is rolling out its new instrument separation models–meaning, we have better stems for you!
The latest improvements to our stems were made possible by better understanding the variety of playing styles and expressions of individual instruments. After expanding our engineering and research team and training a completely new architecture for months, our stems now offer:
- Reduced bleeding from other instruments into the source stems, meaning cleaner stems overall.
- More defined separation quality across vocal, bass, drum, and “other” stems.
- Snappier drums that retain more of the original transients, resulting in punchier kick drums and sharper cymbals.
- Deeper and cleaner bass notes, as well as improvements on expressive techniques such as slap bass.
- Well-balanced piano tones across a wide range of pianos.
- Vibrant guitar sounds with greatly improved sustain across a variety of effect pedals
AudioShake is deeply committed to building the highest-quality sound separation models in the industry so that they can be leveraged by creators, content owners, and developers. So we’ve made it a priority to make these models available to customers of all sizes, deploying them across AudioShake Live, AudioShake Indie, and our API.
How do the new models stack up to others? Our vocal model is nearly 2dB better in SDR compared to the best open-source model (Demucs), and about 1dB better on average across all sources. (You can find more information on SDR scores here). As the industry leader in music stem separation, we’ve been incredibly fortunate to work with artists, labels, publishers, producers, developers, and distributors of all sizes, and have gotten to work on a number of great music projects this past year, including:
- Remastering De La Soul’s first six albums for digital release
- Mastering Fast X’s soundtrack with Immersive Mixers
- A spatial mix of Nina Simone’s Little Girl Blue
- A full live album of Oscar Peterson
- An ODB sample on SZA’s SOS
We’ve also worked to make our technology available on other platforms–like Chordal for sync licensing; OOONA, Yella Umbrella, and Cielo24 for localization–and to help power third-party applications like the mastering company Masterchannel, AI mixing company Tuney, AR app Minibeats, real-time stem separation on djayPro—and many more.
For audio professionals, the quality difference will be clear. But we won’t stop here. We look forward to continued improvements in quality, speed, and access points in the years to come.