By Steve Crouch.

If academic projects were families, then SoundSoftware would surely be the Software Sustainability Institute's sister. Both projects assist researchers with sustainable development, and while the Institute works with researchers from a range of disciplines, SoundSoftware specialises in supporting audio and music researchers.

After having those pleasant folks from SoundSoftware attend and participate in the Institute's Collaboration Workshop earlier this year, it was a delight to be invited and present at their event SoundSoftware 2012 this week. With over 70 attendees and a wide range of presentation topics, it made for an enjoyable and very engaging day.

An excellent (and slide free) keynote from Cameron Neylon kicked-off the event. Cameron made the insightful point that experimental research and computational research share many similarities and have a lot to learn from each other: isn't test-driven development similar to a clinical trial? In both cases, you need to be able to judge success and failure, so aren't unit tests really like experimental controls?

It was particularly interesting to see themes that were raised at our the Collaboration Workshop iterated here in the context of music and audio research. Reproducibility of research was covered emphatically throughout the day: from software development techniques that assist reproducibility, to using workflows to capture and reproduce Music Information Retrieval experiments (as presented by Dave De Roure, one of the Institute's co-Investigators). Cameron's keynote summed up the situation perfectly: the current approach to publishing research results is that of a gallery - put it out there, and say isn't this beautiful. We need to start publishing the substance of our research, which can then be used to reliably reproduce the results. And, of course, this is issue is common to all fields of research.

Very much related to reproducibility, the importance and methods of sharing data was often raised. Steve Wellburn talked about the value of well-defined and well-followed data-management plans, and discussed their increasing importance in securing funding. The importance of management plans for software arose later in the day, again echoing similar discussion at the Collaboration Workshop.

A fascinating and very enjoyable talk from Bob L. Sturm looked at the issues with the Tzanetakis music genre dataset, which is very widely used in music research as the corpus to automatically categorise music by genre. The audience was asked to classify various dataset clips by genre, which was not only fun, but informative: the Tzanetakis dataset had some quite odd classifications. Would you really classify the Willie Nelson track Uncloudy Day as Classical?

The Software Sustainability Institute would like to thank SoundSoftware for the invitation to attend and present at this informative and insightful event, and for the opportunity to see the how challenges in software sustainability are being addressed in the field of audio and music research.