What Is Source Separation?

Source separation is the distinction of multiple sources in a signal to make it possible to pick one or more out while discarding the others. This has important implications for the development of hearing aids that will allow people to distinguish different noises in an environment like a party or a train station. The more sources present, the harder it can be to clean up the signal to achieve meaningful data. Applications including speech recognition software also use source separation to provide more utility to users.

This phenomenon is illustrated by the so-called "cocktail party problem." People in a crowded room often have difficulty separating out the myriad sounds of noise, including people talking, musicians, footsteps, and other sources of sound. For individuals with good hearing, it is possible to drill down and focus on a particular voice, such as a single speaker. People who are hard of hearing may have trouble with source separation, and may need assistance like hearing aids to navigate crowded, noisy spaces.

Hearing aids do not just turn up the volume to make everything more audible. They also submit signals to some auditory processing before routing them into the ear. They need source separation technology to pull apart the different sounds in a room and determine which the listener is most likely to want to be able to hear. The voice of someone nearby should rank higher than a conversation in a different part of the room, for example.

If hearing aids are not well programmed, they can be very uncomfortable to wear. They may provide a jumble of noise without a single meaningful signal, and can make it impossible to people to hear what is going on around them. Bad signal processing may prompt people to turn their aids off for comfort, which defeats the purpose of wearing them in the first place. The development of advanced source separation technology allows for more precision in hearing aid design to pull out voices and step down other sounds.

A variety of source separation algorithms can be used on not just hearing aids, but other auditory processing devices. Speech recognition systems need to be able to pull voices out of background noises, for instance. Musicians work with source separation to clean up recordings. Restoration of old recordings can also involve the processing of the signal to pull out meaningful sounds, like a trumpet player in a jazz band, and silence unwanted noise, like a waitress dropping a glass.