2Hz Launches NVIDIA GPU-optimized Noise Suppression Technology at GTC 2019

We are happy to announce the general availability of our GPU-based Voice Library optimized for large scale deployments inside server infrastructure (cloud services, data centers, etc).

Figure 1: The New NVIDIA T4 GPU

When it comes to running AI based voice enhancement workloads (such as noise suppression) in the cloud, the main challenge is:

How do we scale it to 100K+ concurrent streams in a cost-efficient way and meet the latency constraints for real-time communications?  

Fortunately, NVIDIA is innovating in this space rapidly and has launched an important product lately – NVIDIA T4 GPU.

T4 (similar to V100) GPU is based on Tensor Cores and enables high-performance computing while still being highly power-efficient and cost-effective.

The right diagram below shows how many concurrent streams different GPUs can process with 2Hz Noise Suppression technology:

Figure 2: T4 can process 2300 concurrent streams and V100 can process 2x more. 
P100 is 2x more expensive and way less power-efficient.

The left part of the diagram shows how well 2Hz technology performs against 3GPP voice quality and noise suppression benchmarks. 

This is an amazing result given the price and power-efficiency of T4 GPU.

“Network and commercial service providers have been eager to deploy our solution to improve the voice calling experience across their customer base. NVIDIA T4 and V100 GPUs make it possible to deploy such technologies cost-effectively at scale”

Davit Baghdasaryan, CEO of 2Hz

Key Features of 2Hz technology when deployed in the cloud:

  • Unsurpassed voice quality
  • Low latency
  • Supports bidirectional traffic
  • Codec and transport independent
  • End device and call originaton independent

Early Access

Please leave your details below if you would like to evaluate our Voice Library.