The task described here is that used in the PLoS ONE paper entitled "Continuous Robust Sound Event Classification Using Time-Frequency Features and Deep Learning". It builds upon the long established standard isolated sound evaluation task first described by Jonathan Dennis, and widely used by a number of other authors - for evaluating classifiers of isolated sounds.By contrast the current task extends this to potentially overlapping sounds with no a priori knowledge of start and end points. The advantage of having a standard task is obvious: it makes experiments easily repeatable by others, and eases the comparison of results when other authors make use of the same method to evaluate their own research. With at least 15 state-of-the-art sound classification papers published with Dennis' isolated sound event detection method, it is easily the most popular task defined to date.In this current task, exactly the same raw material is used, but is extended through protocol and setup into a continuous, overlapping and robust classification task. The task uses freely available sound recordings from the Real World Computing Partnership (RWCP) Sound Scene Database in Real Acoustic Environments. These must be obtained directly by the RWCP, and are free for non-commercial or academic users, while commercial users are charged a small free.Robustness evaluation is performed by mixing raw sounds with background noises from the NOISEX-92 database at several signal-to-noise (SNR) levels. The NOISEX-92 data is widely available online for download.