Real time software-based Voice Cancellation is the technology of removing, in real-time, ambient voices of other people talking in the background. This technology is mainly required in verticals like call centers and open office in which people are co-located and usually sitting relatively close to each other and disturbing each other. The Voice Cancellation is a unique field in the arena of noise cancellation technologies. As opposed to the common noise cancellation technologies that simply need to distinguish between human voice and non-human sounds (like car horn and typing sounds) the voice cancellation technology needs to IDENTIFY the origin of each human voice in order to decide if it belongs to the primary speaker or not.

Now, that we have highlighted the difference between voice cancellation and noise cancellation, let’s discuss three possible methods that can enable real-time software-based voice cancellation.

Registering a Voice-Print

In order to enable this method, the speaker needs to provide a one-time register of his/her voice-print. Afterwards, during the calls the voice cancellation will try to match the voice to the voice-print and only matching voice will be considered as a legitimate part of the call. There are two main disadvantages to this method:

1. The accuracy of this method is not good enough since it will not be able to distinguish in real-time between similar voices. Surely there are known technologies, mainly for security purpose, that have a very accurate ability to distinguish between voices but they do not work in real-time and therefore are not applicable for real-time applications.
2. It requires a prerequisite registration phase making it cumbersome.

Use Volume difference

In this method, the Voice Cancellation algorithm will assume that the speaker is using headsets and his/her voice is picked up significantly stronger than the voices of other people in the background. This method will remove any voices that are relatively low. There are two main disadvantages to this method:

1. It cannot cancel loud voiced picked up by the headsets.
2. If it is too aggressive it might also cancel low voices of the primary speaker.

Building a Noise Map

In this method, the Voice Cancellation algorithm picks up the voices coming from multiple locations and correlates them. As a result it can build a Noise Map of the call center or open office and based on this Noise Map identify the exact location/origin of each human voice. The outcome is that any voice that is not originated from the location of the primary speaker will be cancelled even it is very strong or very similar to the voice of the speaker. For more information you are invited to take a look at our Noise Firewall solution that is implementing this method.