2024Prime Minister's Award, Korea Invention Patent Exhibition
Awarded for multi-microphone speech processing technology related to mpBeamforming.
About mpWAV
We build AI voice interfaces you can trust in the real world
mpWAV is an AI voice interface technology company that helps products listen to, understand, and respond to users in real environments where ambient noise and device echo exist.
We develop the full voice stack — single- and multi-microphone enhancement, wake-word detection, sound source localization, multi-party voice processing, speaker diarization, on-device speech recognition, and a conversational language model.
Beyond algorithms, we provide what real implementation requires: microphone arrays, multi-channel audio I/O, DSP·FPGA·AP porting, and product integration.
Who We Are
What kind of company is mpWAV?
mpWAV is an AI voice technology company that improves speech recognition and voice interface performance in real-world noise.
We connect preprocessing that refines the audio arriving at the microphone with wake-word, localization, and speaker analysis, on-device speech recognition, and conversational AI — and implement it all in the customer's software and hardware environment.
We don't stop at cleaning the audio — we connect it so the product understands the user and responds.
Company at a Glance
From research to product — applying voice technology to reality
Founded
2020
mpWAV was founded to solve the speech recognition problems that occur in real product environments.
Business
AI voice interface SW·HW
We develop voice enhancement, voice interaction analysis, speech recognition, conversational AI, and multi-channel hardware.
Technology portfolio
10 voice technology families
mpAEC · mpBeamforming · mpAB · mpNC · mpSeparation · mpDiarization · mpLocalization · mpWWD · mpASR · mpLLM
Product implementation
Software + HW + Embedded
From software validation to microphone arrays, DSP·FPGA·AP porting, and SoC expansion.
Key applications
Robots · Kiosks · Vehicles · Smart Devices · Meetings · Industrial Acoustics
We compose the required technology stack per product and site.
Team
11 people
Specialists in speech signal processing, AI, software, embedded systems, and hardware. (Per official company material, June 2026)
CEO
Hyung-Min Park
Leads mpWAV with research and technology commercialization experience in speech, language, and signal processing.
The Problem We Solve
Voice AI advances — but real products face different problems
Speech recognition models are advancing fast, but a real product's microphone never receives only the user's voice.
Conversation, music, road noise, motor noise, the product's own speaker echo, and overlapping speech all arrive together.
Some products cannot fit multiple microphones — and after recognition, the product still needs the user's position, the speaker's identity, intent, and dialogue context.
mpWAV technically connects the entire flow from input audio to the product's final action.
Ambient noise & far-field speech
The user's voice can be buried under conversation and household or machine noise.
Product speaker echo
Prompts and responses played by the product loop back into the microphone.
Single-microphone constraints
Earbuds and small devices often cannot fit a multi-microphone array.
Wake-up and user position
Robots and vehicles must know when — and from which direction — the user spoke.
Multiple speakers
Meetings and voice chat require separating overlapping voices and speakers.
Context and intent
Kiosks and robots must connect recognized sentences to product functions and actions.
Voice interface performance is decided not by the ASR model alone, but by input, analysis, recognition, dialogue, and product integration together.
AI Voice Technology Stack
From voice input to recognition and dialogue
Speech Enhancement
Improving voice input
- mpNC
Reduces ambient noise from a single microphone.
- mpAEC
Reduces acoustic echo from the product's speaker.
- mpBeamforming
Reduces ambient noise and strengthens the target voice with multiple mics.
- mpAB
Integrates mpAEC and mpBeamforming to handle echo and noise together.
Speech Refinement
Speech separation and speaker diarization
- mpSeparation
Separates multi-party or overlapping voice input by speaker.
- mpDiarization
Separates who spoke when in multi-party audio.
Speech Recognition & Dialogue
Wake-word detection, recognition, and dialogue understanding
- mpWWD
Detects the wake word and activates the voice interface.
- mpASR
On-device end-to-end speech recognition converting voice into text or product commands.
- mpLLM
Analyzes intent and context of recognized sentences, connecting them to product functions and dialogue flow.
- mpLocalization
Estimates the speaker's direction or sound source position.
Product Implementation
Implementation foundation
- Multi-channel microphone arrays
- Speaker output & AEC reference
- DSP·FPGA·AP
- Edge AI
- Semiconductor IP·SoC
Why mpWAV
Integrated voice interface technology built for real products
Noise control that preserves the target voice
If noise reduction damages the user's voice along with the noise, recognition performance can drop.
mpWAV designs its technology to reduce ambient noise while minimizing target-voice distortion and recognition loss.
Less repeated per-product tuning
We optimize from the actual input signals rather than fixed microphone position data, reducing the repeated tuning burden when microphone configurations change.
Single- and multi-microphone coverage
Products with an array use mpBeamforming and mpAB; small products that cannot add microphones use mpNC.
From input to recognition and dialogue
Beyond preprocessing, we connect wake word, localization, speakers, ASR, and LLM as one voice interface architecture.
Works with your existing ASR
You don't have to replace your recognition engine — mpWAV preprocessing applies in front of it.
Software and hardware together
Beyond algorithms we cover microphone arrays, audio I/O, and FPGA·DSP·AP porting.
| Product development challenge | mpWAV approach |
|---|---|
| Voice damaged during noise removal | Preserve target voice, consider ASR results |
| Re-tuning when microphones change | Auto-optimization from input signals |
| Single-microphone product constraints | mpNC single-mic coverage |
| Preprocessing only | Wake·location·speaker·ASR·LLM connected |
| Full ASR replacement burden | Applies in front of existing ASR |
| SW and HW developed separately | Arrays and porting integrated |
From Research to Product
Connecting research results to real product voice experiences
mpWAV's technology is grounded in long-term research in speech and audio signal processing and speech recognition.
Technology validated in papers and patents expands into software, AI models, FPGA·DSP implementations, microphone array modules, and industrial PoCs.
- 1
Research
Speech/audio signal processing and noise-robust speech recognition.
- 2
Patent
Core algorithms and implementation technology protected as domestic and international IP.
- 3
Software·AI
Implemented as voice enhancement, interaction, recognition, and conversational AI.
- 4
Embedded·Hardware
Transferred to real products via FPGA·DSP·AP and microphone arrays.
- 5
Industrial PoC
Validated with real product data and usage environments.
- 6
Product·Scale-up
Expanded as software, modules, licenses, and semiconductor IP.
Research → Patent → Algorithm → Embedded System → Product
- Multi-channel acoustic echo cancellation
- Multi-microphone beamforming
- Speech separation and restoration
- Robust ASR preprocessing
- Real-time, on-device speech recognition
- Multi-channel audio I/O preprocessing hardware design
Applications
The right voice technology for each product and site
Robots
Wake-word detection, user direction estimation, motor/fan noise and robot speaker echo processing, connected to commands and dialogue.
- mpWWD
- mpLocalization
- mpAB
- mpASR
- mpLLM
Kiosks
Captures far-field voice in store noise, recognizes menus and options, and connects to conversational ordering.
- mpAB
- mpASR
- mpLLM
- Mic-Array HW
Vehicles & mobility
Handles wake word, seat/speech direction, driving noise, and car audio echo.
- mpWWD
- mpLocalization
- mpAEC
- mpBeamforming
- mpASR
Smart devices & earbuds
Single-microphone enhancement and on-device commands for products that cannot add microphones.
- mpWWD
- mpNC
- mpASR
- mpLLM
Meetings & voice chat
Processes multi-party or overlapping voices and structures per-speaker utterances and meeting content.
- mpAEC
- mpSeparation
- mpDiarization
- mpASR
- mpLLM
Factory acoustic inspection
Analyzes normal and abnormal acoustic patterns of motors and equipment amid complex production noise.
- mpAEC
- mpBeamforming
- mpAB
- Anomaly Detection
Hearing assistance
Applies voice enhancement to everyday conversation and public hearing welfare through ClearSense Audio.
- mpNC
Implementation Experience
Validated in real products and real sites
Robot ASR preprocessing
Validated speech recognition preprocessing for home robots, showroom robots, and care robots.
Kiosk module
Deployed an ASR preprocessing module on kiosks with a microphone array and multi-channel audio I/O.
Mobility PoC
Tested the voice interface in a moving vehicle, with road noise and cabin speaker echo arriving together.
Factory anomaly detection
Validated detecting abnormal motor sounds amid complex production noise without a separate test chamber.
Hearing assistance field trial
Ran a 100-participant field trial of ClearSense Audio at the Nonhyeon and Bangbae senior welfare centers, scoring 5.8 out of 7 for satisfaction.
Engineering Capabilities
We run the algorithms inside real products
Software·SDK
Integrates mpWAV technology into existing systems or in front of ASR.
AI Model
Fits mpWWD, mpASR, and mpLLM to the target product and domain.
Mic-Array HW
Designs multi-channel microphones, speaker output, and AEC reference structures.
DSP·FPGA·AP Porting
Optimizes for real-time execution on the product's compute platform.
License·Co-development
Designs collaboration structures around per-product data and requirements.
Semiconductor IP·SoC
Expands into dedicated structures for volume production and miniaturization.
Research-Driven Team
A team with 25+ years of research and product implementation capability
mpWAV is a specialist team connecting speech and language research, AI models, software, embedded systems, and hardware development.
CEO Hyung-Min Park earned his B.S., M.S., and Ph.D. in Electrical Engineering at KAIST, with research and teaching experience at Carnegie Mellon University's Language Technologies Institute and Sogang University.
- 25+ yrs
- Speech, language & signal processing research
- 70+
- International journal papers in speech and signal processing
- 40+
- International conference papers
- 50
- Cumulative patents of the core researchers
- 16
- Technology transfers to industry
- 60
- Academic, industrial & government research projects
Leadership
Hyung-Min Park
CEO
Ph.D., KAIST · Professor, Electronic Engineering & AI, Sogang University
Byung Joon Cho
CTO
Ph.D., Sogang University
Ui-hyeop Shin
CAIO
Ph.D., Sogang University
Jun Hyung Kim
COO
Ph.D., Sogang University
Heeman Kim
CSO
M.S., KAIST
Recent publications
Our work on speech separation and speech restoration appears at the top machine-learning venues. Ui-hyeop Shin, our CAIO, is first author on both.
- U.-H. Shin, S. Lee, T. Kim, and H.-M. Park, “Separate and reconstruct: Asymmetric encoder-decoder for speech separation,” NeurIPS 2024
- U.-H. Shin, J. Ko, W. Jeong, and H.-M. Park, “Query-based asymmetric modeling with decoupled input–output rates for speech restoration,” ICML 2026
Intellectual Property
Core technology protected by domestic and international IP
mpWAV's corporate intellectual property holdings:
Korean patents
7 registered · 3 pending
International patents
7 registered · 10 pending
Trademarks
2 Korean · 2 international
Design rights
4
Awards and Certifications
Research results and competitiveness, confirmed by awards and certifications
2024
2025NET New Technology Certification
Certified by the Ministry of Trade, Industry and Energy for real-environment voice interface preprocessing.
2026
Innovative Product for Public Procurement
ClearSense Audio was designated an innovative procurement product, establishing a base for public sector deployment.
2024CES 2024 Innovation Awards, two categories
Recognition for ClearSense Audio, which applies mpWAV voice enhancement to everyday listening.
Milestones
Expanding from research to products and markets
2020.11
Incorporated
2020.12
Eight patents transferred from Sogang University
The foundational algorithm patents moved from Sogang University to the company.
2021.12
2nd place, 2021 AI Grand Challenge Phase 3
2022.12
Winner (Minister's Award), 2022 AI Grand Challenge Phase 3
2023.05
Venture Enterprise certified
2023.11
Double winner, CES 2024 Innovation Awards
Honored in both Mobile Devices, Accessories & Apps and Digital Health.
2023.12
Winner, AccessABILITY Awards by USA TODAY Reviewed
2024.07
Winner, 18th Korea Excellent Patent Awards (Electronics/IT)
2024.12
Winner, Prime Minister's Award, Korea Invention Patent Competition
Awarded for the mpBeamforming patent — microphone-array noise removal.
2024.12
Registered in Venture Nara, Korea's public procurement platform
2025.04
Selected for Deeptech Incubator Project for Startup 1000+ (DIPS)
For an all-in-one on-device stack spanning preprocessing, recognition, dialogue, and synthesis.
2025.05
NET (New Excellent Technology) certified
Certified as preprocessing technology for speech recognition and enhancement in conversational interfaces.
2025.09
Grand Prize, AI Innovation Procurement Expo
2025.11
AIoT Innovation Awards, President's Award of IITP
2026.03
Designated an Innovative Product for Public Procurement
ClearSense Audio received the designation from Korea's Public Procurement Service.
Business Model
Collaboration models for every product and development stage
Software·SDK
Individual technologies or the full voice interface stack applied to your product.
AI Model
mpWWD, mpASR, and mpLLM fitted to your product and domain.
Microphone array module
Multi-channel voice input hardware designed for robots and kiosks.
DSP·FPGA·AP porting
Implemented to run on your product's compute platform.
Technology licensing
Voice enhancement, interaction, recognition, and industrial acoustic analysis.
Joint development
Microphones, preprocessing, ASR, LLM, and product APIs connected in one project.
Semiconductor IP·SoC
Dedicated voice interface structures for volume production and miniaturization.
Global and Technology Roadmap
Real-environment voice technology, to more products and global markets
Building on voice interfaces for robots, kiosks, and mobility, mpWAV is expanding into smart devices, meetings and voice chat, and hearing assistance.
The following is a roadmap the company is pursuing — not completed results.
1
Core Applications
Robots · Kiosks · Mobility
Strengthening preprocessing and product integration for real noise environments.
2
Expanded Applications
Smart Home · Hearing Assistance · Meetings · Voice Chat
Applying single-mic technology, speaker analysis, recognition, and conversational AI to new product lines.
3
Global Products
International business and language expansion
Scaling language-independent preprocessing and on-device models into global products.
4
Semiconductor Platform
Semiconductor IP · dedicated SoC
Pursuing semiconductor collaboration for volume production and miniaturization, built on FPGA·DSP implementations.
Our Vision
So people and products can converse naturally — even in the real world
The voice interface mpWAV pursues is not technology that only works in a quiet lab.
Amid conversation, music, vehicles, and machine noise, products must hear the user's call, know their position and identity, and understand intent and dialogue context.
mpWAV connects input enhancement, interaction analysis, on-device recognition, and conversational AI to deliver voice experiences you can trust in real environments.
Company Information
mpWAV company information
| Company | mpWAV Inc. |
|---|---|
| Korean name | (주)엠피웨이브 |
| CEO | Hyung-Min Park |
| Founded | 2020 |
| Business | AI voice interface software & hardware |
| Technology | Voice enhancement · Interaction · ASR · LLM · Multi-channel HW |
| [email protected] | |
| Phone | +82-2-705-8916 |
| Headquarters | Teilhard Hall 405, 35 Baekbeom-ro, Mapo-gu, Seoul, Korea |
FAQ
Frequently asked questions about mpWAV
mpWAV is an AI voice technology company that improves product voice interfaces in real-world noise.
We develop voice enhancement, wake-word detection, localization and speaker analysis, on-device speech recognition, a conversational LLM, and multi-channel hardware.
Noise and echo reduction are part of our core technology.
The full portfolio spans mpNC, mpAEC, mpBeamforming, mpAB, mpSeparation, mpDiarization, mpLocalization, mpWWD, mpASR, and mpLLM.
Yes. mpNC is built for those products.
It handles single-microphone noise control for earbuds and small smart devices that cannot fit an array.
Quantitative performance and supported platforms should be validated under real product conditions.
Yes — we build both ourselves.
mpASR, an on-device end-to-end speech recognition technology, and the conversational mpLLM.
Actual runtime performance depends on model version and target platform.
Yes.
mpAEC is published in IEEE Transactions on Signal Processing and mpBeamforming in IEEE/ACM Transactions on Audio, Speech and Language Processing. Our speech separation work appeared at NeurIPS 2024.
The company holds 7 registered and 3 pending patents in Korea, and 7 registered and 10 pending abroad.
Both.
mpWAV connects voice algorithms and AI models with microphone arrays, multi-channel audio I/O, and FPGA·DSP·AP porting.
Per official company material (June 2026): 7 Korean patents registered and 3 pending, 7 international patents registered and 10 pending.
No.
The 50 patents are the cumulative record of the core researchers, listed separately from mpWAV's corporate patents.
We have deployment or validation experience in robots, kiosks, mobility, factory acoustic inspection, and hearing assistance.
Smart devices, meetings and voice chat, and defense are expansion areas built on our existing technology.
Yes — both are available.
Depending on the product and project: software, AI models, microphone arrays, porting, technology licensing, joint development, and semiconductor IP collaboration.
Work with mpWAV
Build AI voice interfaces that work in the real world — with us
Tell us your product type, microphone and speaker structure, dominant noise, required voice features, and target platform — we will review the right mpWAV technology and collaboration model.
From single-mic noise control to multi-channel echo and noise processing, wake·location·speaker analysis, on-device ASR and conversational LLM, microphone arrays and DSP·FPGA·AP porting — start at whatever stage your product is in.