Case studies
Validated in robots, kiosks, vehicles, factories, and care settings. These are product development and module supply engagements, not research demos.
Customer names and logos appear only after we have written consent from each company. The cases below describe the technical work without identifying the customer. Under NDA we can walk you through the specifics.
Third-party verification
TTA test report TTA-25-1103
ClearSense Audio v1.1.0 · tested 2025-06-09 to 12 · issued 2025-11-17
100 utterances at 0 dB input SNR (SI-SDR, SNR) · 100 utterances at 5 dB real-world SNR, 65 dB speech and 60 dB noise (noise reduction). Corpora: LibriSpeech ASR and AIHub Korean speech.
SMBA Didimdol TTA test report
Subject: mpAB (mpAEC + mpBeamforming combined)
65 dB speech and 60 dB noise. SRAM requirement 0.571 MB.
Deployments
Voice interface preprocessing for a home appliance robot line
Array-free adaptive preprocessor
Care robot microphone array with integrated preprocessing module
Multi-channel audio I/O and ASR preprocessing module
Acoustic inspection on the production line
Everyday conversation support in noisy places
In-cabin speech preprocessing
Joint research with ETRI, a leading government-funded research institute in Korea
mpAEC echo cancellation · mpBeamforming multi-microphone noise removal · mpNC single-microphone noise suppression · mpAB combined · mpSeparation source separation See the technology
Customer stories
These are the cases we can describe in full, with written consent from the customer. Unlike the anonymous summaries above, they carry the environment, the problem, and the outcome end to end.
We are preparing cases we have consent to publish. Until then, please refer to the anonymous summaries and the third-party test reports above.
Public pilot
We validated the effectiveness and user satisfaction of the ClearSense Audio smart listening app at two senior welfare centers in Seoul.
The project was selected for the Seoul Business Agency programme for technology that serves people with reduced access.
How a project starts
Environment analysis
We map the kinds of noise, the microphone count and placement, and the compute platform together. You can start without field recordings.
Data review
We take your own recorded speech and separate out what is actually causing recognition to fail.
PoC and validation
We compare before and after on that data. The numbers are re-measured against your environment.
Product integration
We deliver it as a library, a module, or a DSP or FPGA port — whichever fits your stage.
FAQ
Because we publish a name only after we have that company's written consent.
Speech preprocessing goes into the pre-launch development stage of a product, so the supply relationship itself is often confidential to the customer. Under an NDA during a consultation we can walk you through the relevant cases in detail.
The numbers we publish are the ones an accredited lab measured.
This page carries the measured values from TTA test report TTA-25-1103 and the SMBA Didimdol report, together with the test conditions. Values measured inside an individual customer's environment belong to that customer, so we do not publish them.
It varies by environment, so we measure it on your own data.
The kind of noise, the microphone count and placement, the distance to the speaker, and the compute platform all drive the result. That is why every project starts with a PoC on customer data.
No. It keeps working without retuning.
Instead of taking the relative microphone positions as input, mpBeamforming optimizes itself from the incoming signal. Across 2 to 8 microphones you can change the count, the layout, and the industrial design without re-fitting the algorithm.
Tell us your product type and noise environment and we will put together the nearest case and how it would apply.
