Noise removal in a call center environment is a challenging task. As opposed to non-human ambient noise like street noise and car horn, the background noise in call centers is actually human voices of people talking in the background. When performing noise removal in a contact center environment the challenge is to identify which of the human voices are a legitimate part of the call and belong to the primary agent and which of the human voices belong to agents in the background and therefore should be treated as noise.

The noise identification challenge

If you listen to one second of a call that is coming from the street, you will be able to immediately identify if this one second contains noise (e.g. car horn) or human voice. On the other hand, if you listen to one second of a call that is coming from a call center, you will surely hear a human voice. How will you be able to decide if this human voice belongs to an agent in the background or belongs to the primary agent? The answer is that you probably won’t be able to make this decision by just listening to 1 second. Noise removal algorithms have even a harder time since they do not have the luxury of listening to a full second before making a decision. Their decision must be taken within a fraction of a second.

How can we overcome the challenge?

Now that we understand the challenge, we need to find a good solution for it. What if, in addition to hearing one second of the call, you will also hear one second of the call that is made by the agent that is sitting on the left and one second of the call made by the agent that is sitting on the right? So, now you have a total of three seconds that you can listen to in order to make a decision if the content is noise or not. It is clear that with this additional information you have a much better chance to decide if the voice in the original one second belongs to the primary agent or if it belongs to an agent in the background.  Moreover, in case you identify it as a background noise, you could probably also tell if it is coming from the left or from the right. As you can see, by adding additional reference information we are turning the original impossible challenge to a realistic one. This technique is called Reference-Based Noise Reduction (RNR).

Noise Reduction Headsets, the new generation

Back to the original title of this post, the next generation of noise reduction headsets for call centers should utilize this advanced technology and work not only on a single agent, but should have the capability to receive and correlate audio that is coming from multiple agents and use this reference data to identify the ambient voices in the call. Such multi-agent headset would be able to effectively attenuate the ambient voices in the contact center.

Multi-agent headset

As of today, such multi-agent headsets do not exists and the headsets are working stand-alone. Fortunately, the “multi-agent headset” technology is available today in a software product. The Noise Firewall is an innovative software solution that collects the audio from multiple agents, correlates the multiple audio streams, identifies the ambient noise and removes it. In addition, the Noise Firewall provides advanced visibility. For example, it builds and displayed an interactive Noise Map of the call center and a centralized real-time statistics and alerts on call quality.