Environmental Noise Cancellation voice processing software package providing bi-directional enhanced speech clarity and intelligibility for resource-constrained devices

Clear speech is fundamental to communications and has become ever more critical in our hybrid, distributed world, whether placing calls on TWS earbuds, talking to teammates on gaming headsets, or even sending voice notes on a smartwatch. Regardless of the conditions around you or your communication partner, the primary goal is to be understood.

ClearVox ENC is an embedded, edge-inferencing Neural Network-based algorithm for Environmental Noise Cancellation (ENC). The sophisticated algorithm is tuned to provide enhanced speech intelligibility in challenging environments on single-mic, resource-constrained devices like TWS, headsets, wearables, and Bluetooth speakers.

Traditional ENC solutions only improve the outgoing speech in a call, benefiting the downstream party. ClearVox ENC uniquely provides bi-directional improvement, so both parties hear each other loud and clear through both transient and stationary noises.

Benefits

Clear, intelligible speech for both parties on a call regardless of what system the downstream partner is using
Fully-embedded inferencing ensures low-latency and high performance without privacy concerns or lock-in to a cloud ecosystem
Optimized NN model with small memory and compute footprint allows ClearVox ENC to run side-by-side with other must-have features even on resource-constrained devices

Main Features

  • Promotes human speech intelligibility in the presence of both transient and stationary noises
  • Bi-directional functionality, cleaning noise from outgoing speech as well as the incoming speech, providing benefit for both parties on a call
  • 16KHz sample rate provides clear communication without overburdening system resources
  • Appropriate for voice communication applications, including hearables, wearables, and smart home devices
  • ClearVox ENC is available today for Ceva’s DSPs: Ceva-BX1, Ceva-BX2 and SensPro2 family, as well as Arm CPUs

Figure 1 – Outbound ENC

 

Figure 2 – Inbound ENC