Training listeners how to judge sound quality in a reliable way is a high priority for Harman’s Director of Research, Sean Olive. A version of the company’s “How to Listen” software has recently been made available in a public beta version, to enable audio enthusiasts to improve their listening skills. It can be downloaded free from http://harmanhowtolisten.blogspot.com/ in either a MacOS or a Windows XP version. It comes together with four high quality music extracts provided by Bravura Records, which can be used as programme material in the listening exercises. A clear instruction manual is also available. Olive argues forcefully that training is a vital part of preparing and selecting listeners when benchmarking and testing audio products in research or commercial settings. His work so far on loudspeaker testing provides evidence that training gives rise to more repeatable and reliable judgements without biasing listeners’ preferences, allowing them to respond in a “language” that is meaningful to engineers designing the loudspeakers being examined. This new software tool is based on the one used by Harman in its own projects, and includes a substantial set of exercises designed to improve one’s discrimination between various peaks and dips in the frequency spectrum. It also deals with high-pass and low-pass effects. As the listener becomes more competent the skill level increases, more demanding tasks are given, and the opportunity is offered to audition the different alternatives when a test is failed. This seems easier to hear with some of the music extracts than others, as they don’t all have substantial energy in the places where the spectral changes are made, but that’s a typical real-world scenario. Considering the difficulties of getting rights-cleared listening material for public consumption, it’s helpful that reasonably useful extracts are provided. An excellent feature is the ability to add your own audio material and use it in practice sessions, where you can set the number of bands and various other aspects of the test. In addition to the frequency band ID training, there is a set of exercises designed to test a listener’s ability to rank extracts on various subjective attributes. This is an area that could be considered a little more controversial, particularly in respect of some of the attributes, but it represents an interesting opportunity to test both your ability to rate subtle differences between sounds and to decide whether you agree with the approach. In this beta version one can rank two or more sounds on attribute scales such as “bright/dull” and “full/thin”. The middle of the scale is marked “ideal”, which makes me a little twitchy because it appears to conflate a descriptive rating with notions of goodness or correctness. What if I want it to sound bright?! In the manual it’s said that “good sound should be neither too bright nor too dull”, but perhaps that’s a matter of opinion. Perhaps “neutral” would be a better label. (To be fair, this system was originally designed for use when testing loudspeakers, which are supposed to reproduce programme material in an accurate or “ideal” way. If the system were to be used for listener training more generally some concepts might need revising.) More acceptable seem the tests that ask the listener to rate the amount of something without introducing a value label into the scale. Try it for yourself and see what you think!