Transcription Is All You Need: Learning to Separate Musical Mixtures with Score as Supervision
This addresses the challenge of data scarcity in music source separation for audio processing applications, offering a novel training approach that is incremental over previous score-informed methods.
The paper tackles the problem of music source separation without requiring large collections of isolated sources for training by using musical scores as weak labels, resulting in improved separation and transcription performance compared to temporal weak-labels and further gains with adversarial losses.
Most music source separation systems require large collections of isolated sources for training, which can be difficult to obtain. In this work, we use musical scores, which are comparatively easy to obtain, as a weak label for training a source separation system. In contrast with previous score-informed separation approaches, our system does not require isolated sources, and score is used only as a training target, not required for inference. Our model consists of a separator that outputs a time-frequency mask for each instrument, and a transcriptor that acts as a critic, providing both temporal and frequency supervision to guide the learning of the separator. A harmonic mask constraint is introduced as another way of leveraging score information during training, and we propose two novel adversarial losses for additional fine-tuning of both the transcriptor and the separator. Results demonstrate that using score information outperforms temporal weak-labels, and adversarial structures lead to further improvements in both separation and transcription performance.