Reference based Noise Reduction

Background

Ambient noise during phone calls is annoying especially when the phone calls are a significant part of your business. Let’s assume you run a call center and there is temporary (or constant) source of noise that is picked up by the phones of the agents and therefore this noise is being heard by your customers. The source of noise can be a nearby coffee corner, other agents shouting, an internal PA system,  etc. Regular noise reduction solutions will find it difficult to automatically identify these sounds as noise since in many cases they contain human voices that in other cases could be a legitimate part of the call.

noise in a coffee corner

Introducing a reference signal

Reference based Noise Reduction (RNR), is a method that enables the noise reduction algorithm to identify which sounds are a legitimate part of the call and which sounds should be considered as noise in the phone call and therefore should be removed.  In order for this technology to work, the noise reduction algorithm should receive a reference audio signal that contains the noise.

capturing reference audio signal

For example such a reference audio signal can be provided by a microphone that is placed in the coffee corner and records the noise coming from the coffee corner. The RNR algorithm can use this reference audio to remove, from the phone calls,  the  noise coming from the coffee corner. Technically, this is done by “subtracting” the audio that appears in the reference signal (i.e. the noise) from the audio of the phone call.

RNR is now available

This sophisticated RNR technology is now available in SoliCall’s products. For example, let’s take a look at the PBXMate product.

Starting from version 1.6.68 of PBXMate, you can easily activate RNR technology. PBXMate allows you to define channels that record/write the reference data (e.g. record audio from the coffee corner) while other channels can use/read this reference data to remove this noise (e.g. phones calls of the agents in call center).

In the above image, “2” is located in the coffee corner and recording the reference noise (i.e. writing the noise to RNR buffer 0). This noise is being used to clean the audio of agents coming from “3” and “10”. In other words, when the algorithm cleans the audio of both these calls (“3” & “10”), the reference noise in RNR buffer 0 is being read and removed from the call.

[EDITED] For the latest updates on RNR implementation in contact centers, please visit PBXMate for call centers.

Conclusion

RNR is a powerful technology that enables you to attenuate from your calls any type of noise including voices of other people talking around you. In order to enjoy this powerful technology you need to provide a reference signal that will enable the RNR algorithm to identify which voices/sounds should be considered as noise and should be attenuated. In addition to PBXMate, this algorithm is also available in our SDK products. For more information please contact us.