A Better Approach
to Signal Processing

Today’s devices use custom hardware to try to mask the noise. This means they need early-stage integration into product design that can delay schedules, limit flexibility and inflate costs. Making things worse, traditional methods are typically only really effective on constant noise sources from fixed locations.

Cypher software starts with a large database of human speech, which is analyzed by a neural network to identify the unique aspects of human speech. This information is used to create a small, fast algorithm capable of detecting speech and isolating from all types of background noise.

Cyphers solution begins with an enormous database of human speech recordings. This information is processed through a neural network to create a catalog of elements related to human speech.

Step 001


On the phone, we take the incoming audio and break it into very small segments to look for specific sound patterns found in human speech.

Things like frequency, harmonics and attack and decay characteristics can distinguish people from environmental noise.

Step 002


Based on the match, different filters are used to cut out virtually everything else.

Step 003


The final step in the Cypher process is post processing. Here we enhance the voice signal and remove artifacts created in the voice-isolation process. This recreates a balanced, full sound as close to the original speaker as possible.

The Result

Standards-based testing shows up to 99% noise reduction and a 20% improvement in audio quality.

Noise Reduction

With no custom hardware required, compared to major phone manufacturers. For more information, contact us.

  • Cypher Baseline
  • Samsung 4.3x Less
  • LG 3.0x Less
  • Apple 8.1x Less

How To Deploy

Cypher software can be deployed in three different ways: Embedded in Existing Hardware, Integrated into the OS or as an Application running on top of the OS. Cypher only needs input from two microphones in all implementations.

Embedding into hardware

When embedded in the hardware, Cypher is typically deployed in the central processing unit or a DSP used to process audio, speech or baseband.

Diagram of where Cypher can be embedded in the hardware.

Integrating into the OS

Cypher can also be deployed in a phone's operating system using two general options. To illustrate in an Android architecture, Cypher can be stitched into either the media framework at the application framework level or embedded into the audio drivers at the kernel level.

Diagram of where Cypher can be integrated into an architecture.

There are other posibilities and other operating system models to consider. Contact Us with any questions or for specific integrations.

How Fast? How Small?

One part of isolating voice from noise involves splitting the audio stream and taking a very quick look to see if it matches the speaker's voice. In a traditional speech recognition system small delays to process these decisions aren't crucial. But in a voice call, any lag greatly hinders conversation. Cypher is optimized to buffer as little as 24 milliseconds to create a clear signal.

As for size, we have a couple of different platforms. The smallest is a Beagle Board with an ARM Cortex-A8 clocked at 600 MHz, 256 MB RAM and 2 GB NAND. It's loaded with Ubuntu and the Dhrystrone test reports 1,054 DMIPs available before we turn our algorithms. In the most compact configuration, Cypher uses only 4.9% of the CPU (52 DMIPS) — withough any hardware specific code optimization. Cypher is running in C.

We also have demos that can run on an off-the-shelf Samsung Galaxy smartphone. Contact us to arrange a live video conference in-person demonstration.

What Did You Say?

Automatic voice recognition (ASR) software is making its way into many applications and devices. Assistants like Apple's Siri, Microsoft's Cortana and Google Voice let people speak their commands into their devices. Unfortunately, background noise can significantly reduce the effectiveness of theses systems.

With Cypher, noise becomes less of an issue. Cypher can cut the word error rate by 40% in noisy environments. This means fewer garbled instructions and a lower risk of throwing your phone to the ground.