In late 2021 we added a multitrack separation model into Chord ai. The current model is able to perform multitrack separation into 4 stems: bass / vocals / drums / other. We are using a deep learning model architecture called U-Net that we have trained on thousands of both real and also synthetic songs to achieve the highest possible audio quality on each separated stem. When tapping the “Separate” button, your audio track will be processed on-device in a few seconds at most, leveraging latest advances for machine learning inference on mobile. Multitrack separation is free, and you also will be able to export stems as individual audio files if you are a pro user.
We are continuously improving our algorithms including the separation model. We plan to release model updates with increasing separation ability and audio quality. We are also working on more fine-grained separation models, starting with 6 stems separation (bass / vocals / drums / guitar / piano / other) coming soon in 2023.