Audioshake uses AI to break down a mix into stems


Image Credit: Audio shake

With a single audio file, Audioshake uses artificial intelligence to isolate vocals, drums, guitar, bass, and other instruments.

Audioshake’s technology could facilitate remixing, mash-ups, remasters, bleeding removal from multitrack live recordings, grabbing samples and instruments for karaoke or background music. Trained on tens of thousands of high quality rods to learn what each instrument looks like, Audioshake uses AI technology to recognize and isolate different components.

We’ve seen similar technology from Deezer before with Spleeter. While Audioshake admits their product isn’t perfect, it’s the best thing to do when you don’t have access to the original rods. They are clearly doing something right, like earlier this year, Audioshake won the Sony Demixing Challenge above from researchers at Facebook and ByteDance (TikTok’s parent company). Sound separation goes beyond music to improve products such as hearing aids.

Labels and publishers can sign up for a 15-minute web conference demo that will take place tomorrow and over the next two weeks. For musicians, sound engineers, music supervisors and producers who need multiple tracks that you own “stemmed”, contact Audioshake. They will create samples that you can choose to purchase.

Audioshake is now putting technology in the hands of the music industry. Audioshake Live is an on-demand stem creation platform. The demo below shows a program as simple as downloading a song and choosing the stems, before being presented with waveforms that can be cut, soloed or mixed with one click. From there you can listen in the player or download the ripped tracks. Audioshake Live is shaping up to be a subscription platform for labels and publishers who regularly need to create stems. Those interested in Audioshake Live can sign up to receive updates here. Musicians who need to create stems on an ad hoc basis can continue to contact Audioshake.

Source link

Leave a Reply

Your email address will not be published.