2
www.radisys.com
RADISYS SOLUTION BRIEF | Engage AI-based Noise Reduction Ensures Clearer Communication Experience
The Radisys Engage In-Call Assistant – a carrier-grade cloud-based speech analytics service - leverages
AI-powered analytics to eliminate these distractions from any personal or business call or meeting.
Communication service providers and conferencing providers can offer noise reduction as a value-
added service to consumers and businesses across multiple verticals. Automatically, without any user
intervention, this service can ensure that anyone, whether they are a C-level executive conducting an
important board meeting, a doctor conducting a virtual visit from home, or a student attending class
online, will not have to apologize for the loud, barking dog in the background when a delivery driver rings
the doorbell. For contact centers, noise cancellation can mean reduced call times and reduced call errors,
which can create millions of dollars in savings, highly satisfied customers, and a tangible, quick ROI.
Shifting from Basic to AI-Driven Technique for Productive Communication
& Collaboration Experience
Traditional approaches to noise filters are not effective in reducing unique and varying noises. Prior
generations of noise cancellation algorithms are not adaptive enough and almost ineffective for
eliminating many background noises. Today's advanced artificial intelligence-based noise cancellation is
designed to reduce unwanted sound by creating a signal that is identical to the unwanted noise but with
the opposite polarity. The two signals cancel out due to destructive interference. Large sound datasets
that provide a foundation for training, together with the ability of today's technology to improve as it is
used, can address the broad range of environments where voice and video interactions take place better
than ever before.
The high level approach consists of three steps:
1. Data Collection: Generate a large dataset of synthetic noisy speech by mixing clean speech
with noise
2. Training: Feed this dataset to the Deep Neural Network (DNN) on input and the clean speech
on the output
3. Inference: Produce a mask (binary, ratio, or complex) that will leave the human voice and filter
out noise
Click the above images to listen to an actual sample of the Engage AI-based noise reduction in action.
(External Vimeo Link).
