We sincerely thank all the reviewers for their insightful comments to help us improve the paper. T o Reviewer #2. 1. Are multiple sources more beneficial? This is largely due to the fact that domain gap also exists among different source domains. We will reorganize the layout of Figure 1 in the main paper to make it more clear. We thank the reviewer for pointing this out.

Despite significant advancements in deep learning for vision and natural language, unsupervised domain adaptation in audio remains relatively unexplored. We, in part, attribute this to the lack of an appropriate benchmark dataset. To address this gap, we present Synthia's melody, a novel audio data generation framework capable of simulating an infinite variety of 4-second melodies with user-specified confounding structures characterised by musical keys, timbre, and loudness. Unlike existing datasets collected under observational settings, Synthia's melody is free of unobserved biases, ensuring the reproducibility and comparability of experiments. To showcase its utility, we generate two types of distribution shifts-domain shift and sample selection bias-and evaluate the performance of acoustic deep learning models under these shifts. Our evaluations reveal that Synthia's melody provides a robust testbed for examining the susceptibility of these models to varying levels of distribution shift.

Drug-repurposing studies are testing a range of compounds to treat COVID-19, but manufacturers may struggle to meet demand if any of these candidates prove effective against SARS-CoV-2. The pandemic has already strained global supply chains and limited the availability of a number of products, including hand sanitizer and diagnostic test reagents. The raw materials needed to make a new antiviral drug would most likely face similar pressures. But a team led by Tim Cernak of the University of Michigan has used an AI-based retrosynthesis program called Synthia to devise alternative routes to 12 leading drug candidates under investigation. The work appears on a preprint server and has not been peer reviewed (ChemRxiv 2020, DOI: 10.26434/chemrxiv.12765410.v1).

It may look like a video game, but the new computer simulation developed by a team of researchers in Barcelona could one day train autonomous cars to be better drivers. Called'Synthia,' the program creates a virtual city complete with pedestrians, traffic signs and other components of an urban environment, automatically annotated at the pixel-level. This allows for a more efficient method of training AI systems, and can be used to teach them to recognize and behave in response to the less predictable aspects of city driving, like a nearby cyclist or adverse weather. A new computer simulation could one day train autonomous cars to be better drivers. Called'Synthia,' the program creates a virtual city complete with pedestrians, traffic signs and other components of an urban environment, automatically annotated at the pixel-level Researchers hope programs like Synthia can be used to improve the abilities of AI to recognize different objects, to make autonomous driving more reliable.