Parasitic Computing Architecture

Parasitic computing is a patented architecture that implements digital signal processing algorithms in the form of a "device driver" for a particular audio chipset. As an alternative to built-in or external hardware DSP modules (meaning higher cost and power consumption), this architecture allows peripheral audio accessories of a host machine (phone, PC, pad, etc.) to make use of the host's resources (storage and calculation power) to process downstream/upstream audio signals from/to user-layer applications, for free.

​

The abundance of resources on the host machine allows the deployment of algorithms with bigger code sizes and greater calculation complexity. It also opens up the possibility of multiple algorithms working in parallel/series, hence possesses the potential for better performance and room for future upgrades.

Parasitic Comuting Architecture

Patent Number: US 2020-0221223 A1

​Audio Processing Algorithms

Ranging from noise cancellation to voice stylization to sound collection, we offer a variety of algorithms for our customers to choose from. We are always looking forward to working with other algorithm companies to come up with new or better algorithms to further bring convenience and new applications to our users. These are the algorithms that we currently offer:

1. Directional Sound Collecting Algorithm

Our directional sound collecting algorithms, based on beam-forming technology, computes audio signals collected by microphones in a fixed array to generate an output preserving sound from sources within designated directions while attenuating/supressing sound from sources outside.

 

The intensity of attenuation is customizable per user's request and can reach up to 40dB SNRI, i.e. much greater than the effect of any directional microphone components. This algorithm is therefore a powerful tool in directional recording scenarios including streaming, interviewing and so on.

kikaGO.AI is able to cancel out most of the surrounding noise UP TO 90dB! 

kikaGO Noise Cancellation Demontration
beforeNC
AfterNC
 

Demo 1

This is a demo of our products at CES. As you can see in the video, the cell phone with our product is able to detect and understand what our users are talking about under a noisy environment of 80+dB.

Demo 2

This is a demo of our products in a driving scenario. We understand that for a lot of people, using the AI assistant while driving is a pain in the neck. The noise from other cars, your engine, or your passengers talking simply means you cannot get your command through to your phone.

 

The following video is a demonstration of how our products can help increase driving safety and fulfill hands-free voice convenience.

Demo 3

We at kikaGO also tested a few phones with voice assistant functionality to demonstrate the difference between the original microphones on your cell phones and cell phones with kikaGO Noise Cancelling products. The following video is shot in an environment of around 85dB.

 

Our colleagues issued the same command to three different cell phones. Two of them are using the microphone on the cell phone while one of them is equipped with kikaGO Noise Cancelling products. Here is the video:

kikaGO.AI powered smartphone v.s iPhone+Siri v.s. Samsung+Google Assistant comparison

2. Environmental Noise Cancellation (ENC) Algorithm

Our Environmental Noise Cancellation algorithm computes audio signals collected from microphones in a non-fixed array to generate an output preserving sound from sources geometrically close to a designated microphone in the array while attenuating/suppressing sound from sources all around.

​

The intensity of attenuation is customizable per the user's request and can reach up to 25dB SNRI. Wired or wireless audio accessories, such as a headset, can therefore be implemented with at least one microphone pointed to the expected sound source and the other auxiliary microphone(s) implanted elsewhere, collecting environment noise to help the noise cancellation. 

kikaGO Noise Cancelato Before and After Comparison

3. Voice Stylization Algorithms

Aside from noise cancellation, we have also developed other audio processing algorithms. One of our strengths is human speech-related processing algorithms. Here are the types of processing that we can do:

​

  • Resonance

Add light or heavy resonance to the voice, beautify your singing as in karaoke.

  • Tone Changing

Change the tone of your voice! We can make you sound like a girl, an old person, or even a duck!

  • Special Effects

We can add or customize any kind of canned sound effects (laugh, applause, etc.) and make your phone call or stream fun and happy one.

  • Speed Change

Speed up your speech like a rabbit or slow down like a turtle!

  • ASMR*

This is our latest research. We are making it possible for normal headphones to perform like ASMR      headphones.

4. Multi-Channel Audio Sync Algorithm for USB Devices

As traditional recording options go, it is not possible to record multiple channels of sound data from phones. We understand that mixers or PCs need to be used for this kind of applications. Also, post-editing is needed for the recorded audio channels to be in sync. This algorithm provides the possibility for Android phones to record 2 or multiple synced channels of audio data for a more convenient and improved approach to recording.

​

This algorithm is especially useful for mobile phone applications because we understand that the potential applications for Voice AI deployed on phones or a more portable audio data collection terminal is a need on the market. This algorithm has already been applied to detect machinery defects by one of our clients.

​

As for traditional PCs, this algorithm and solution setup will replace traditional mixers with the ability to record multiple audio channels in sync. No more post-editing needed to sync up all your audio channels.

​

With this algorithm, we are also working to solve one of Voice AI's biggest problems right now, which is that Voice AIs on the market generally have a hard time understanding human words when in a noisy environment. We believe that this algorithm is a must-have component for future Voice AI deployment.

Noise/Echo Cancellation

  • Acoustic Echo Cancellation (AEC)

  • Crosstalk Suppression

  • Auxiliary Mic ENC

  • Mic Array Beamforming

  • DNN/RNN - AI Models

Audio Mixer

  • Background Sound

  • Music Ambiance

  • Automatic Background Sound volume Adjustment

  • Canned Sound Effects

  • Audio Quality Restoration*

Audio Watermark

  • Anti-Fraud

  • Leak Tracing*

  • Authentication*

  • Voice Signature*

  • Voiceprint Disguise*

*Currently under development

5. Other Algorithms

Aside from above features, we also integrate and develop additional sound processing algorithms for our clients to choose from. With our unique deployment structure, we can always update and add on new algorithms per clients' need with software updates instead of having to change the entire chipset/ module like traditional DSP solutions. Here are the algorithms that we currently offer: