ENGINEERING NEWS

Engineering Flow

Bryan Pardo develops easy-to-use audio software

Bryan Pardo has a PhD in computer science, spent years as software developer, plays several musical instruments, and has mixed several albums. But even to him, audio production software interfaces are confusing arrays of knobs, buttons, and tools that are difficult to master.

“We’re working to turn that completely on its head and make really simple, really intuitive interfaces,” says Pardo, associate professor of computer science at Northwestern’s McCormick School of Engineering and Applied Science. 

With current tools, Pardo says, “People who would have and could have produced something that is great, don’t…We want to make tools that they see the tool and they go, oh, I can do this, and then they do, and then great stuff is made.”

Bryan PardoTo this end, Pardo and his Interactive Audio Lab develop audio production software that uses crowdsourcing to turn complicated production actions into easy-to-understand interfaces and terms.

Tools his lab has developed include Reverbalize, a reverberation tool whose control interface is a word map of terms regular people use to describe reverberation (like “crisp” or “underwater”). The tool learns the right words to map onto particular reverb effects by gathering input from the hundreds or thousands of people on the web who play the sound and give a one-word description of the effect. The most common description becomes the word used. 

When this is done for hundreds of reverberation effects, the resulting map of words forms an easy-to-use interface that has been shown to be easier for non-experts than the existing tools. One simply clicks on the word that describes the desired effect.

Another tool, SocialEQ, is an equalizer that lets the user achieve a desired effect by listening to the sound and rating alternatives. The system modifies the sound with an equalization setting, and the user rates how close that setting is to the desired sound. The system then tries again, with the user rating the result. Most users achieve their desired setting with as few as eight rated examples. Results are then saved to the web to speed the interaction for future users.