Resource-constrained stereo singing voice cancellation

We study the problem of stereo singing voice cancellation, a subtask of music source separation, whose goal is to estimate an instrumental background from a stereo mix. We explore how to achieve performance similar to large state-of-the-art source separation networks starting from a small, efficient model for real-time speech separation. Such a model is useful when memory and compute are limited and singing voice processing has to run with limited look-ahead. In practice, this is realised by adapting an existing mono model to handle stereo input. Improvements in quality are obtained by tuning model parameters and expanding the training set. Moreover, we highlight the benefits a stereo model brings by introducing a new metric which detects attenuation inconsistencies between channels. Our approach is evaluated using objective offline metrics and a large-scale MUSHRA trial, confirming the effectiveness of our techniques in stringent listening tests.

Generalizable Error Modeling for Search Relevance Data Annotation Tasks

Human data annotation is critical in shaping the quality of machine learning (ML) and artificial intelligence (AI) systems. One significant challenge in this context is posed by annotation errors, as their effects can degrade the performance of ML models. This paper presents a predictive error model trained to detect potential errors in search relevance annotation tasks for three industry-scale ML applications (music streaming, video streaming, and mobile apps) and assesses its potential to enhance the quality and efficiency of the data annotation process. Drawing on real-world data from an extensive search relevance annotation program, we illustrate that errors can be predicted with moderate model performance (AUC=0.65-0.75) and that model performance generalizes well across applications (i.e., a global, task-agnostic model performs on par with task-specific models). We present model explainability analyses to identify which types of features are the main drivers of predictive performance. Additionally, we demonstrate the usefulness of the model in the context of auditing, where prioritizing tasks with high predicted error probabilities considerably increases the amount of corrected annotation errors (e.g., 40% efficiency gains for the music streaming application). These results underscore that automated error detection models can yield considerable improvements in the efficiency and quality of data annotation processes. Thus, our findings reveal critical insights into effective error management in the data annotation process, thereby contributing to the broader field of human-in-the-loop ML.