Denoising
Hearing a voice is only part of the challenge.
The rest is keeping it clear over real noise, through the microphone your product has. That's where our engineers help.
Last updated
Problem
Why real noise is harder than the lab
Lab tests prove noise can be removed. Real places are harder, because the noise keeps changing and often sounds like speech.
Noise keeps changing
Traffic, chatter and machines rise and fall while someone speaks.
Noise can sound like speech
Other voices share the pitch and rhythm of the one you want to keep.
Every product sounds different
The microphone, the enclosure and the distance to the speaker all change the sound.
It has to keep up
Speech is cleaned as it's spoken, so there's no time to wait.
| In the lab | In the field | |
|---|---|---|
| Noise | Recorded noise at a set level | Cafés, traffic and machines that change by the second |
| Microphone | A reference microphone | The microphone in your product |
| Speakers | Close to the microphone | Wherever people stand |
| What it proves | Noise can be removed | Voices stay clear where your product is used |
- Understand
Understand your product
We start from your product, where the microphone sits and who speaks to it.
- Tune
Tune for your noise
We adapt denoising to your microphone, your enclosure and the noise around it.
- Validate
Validated with your recordings
We check results with recordings from your own product and site.
Features
Denoising that fits your product
All three come with every use case.
AI speech denoising
A neural network separates the voice from the noise across the whole spectrum.
Works with your audio
From a microphone on the device, or audio your system sends.
Processing on the device
The voice is cleaned on the device, as it's spoken.
Your system: Sound
The voice, and the noise around it.
Your system: Microphone
The microphone in your product.
Our solution: Analysis
Finds the voice in each slice of sound.
Our solution: Denoising
Removes the noise and keeps the voice.
Your system: Your product
Your call, recording or voice control.
Use cases
Different listeners, one approach
A person on a call and a machine listening for commands need different things from the same clean voice, so we develop each use case for its own setting.
Speech denoising
Removes background noise from speech as it's spoken, so calls and recordings stay clear.
Keyword spotting
Cleans the voice first, so spoken commands are recognised over the noise around them.
Evaluation
Interested in this solution?
Get in touch with our engineering team to evaluate it for your system.
Evaluation platform. Firmware and software are under development; each use case requires application-specific validation.
Other solutions
More from our engineers
We take the same approach to arc faults, vision and depth sensing.
