AI Voice Engineering Services
mpWAV applies single- and multi-microphone enhancement, wake-word detection, sound source localization, multi-party processing, speaker diarization, on-device speech recognition, and a conversational LLM to your product.
From PoC on your real audio data through software integration, DSP·FPGA·AP porting, microphone array and multi-channel hardware design, model optimization, licensing, and dedicated SoC collaboration — engineering services matched to your development stage.
Core services
PoC · Software Integration · On-Device AI · Embedded Porting · HW Integration · Licensing · SoC Partnership
Customer Product + Real-World Audio + Target Platform
↓ mpWAV Engineering Services — Technology Selection · PoC · Software Integration
Model Optimization · DSP·FPGA·AP Porting · HW Integration · Licensing · SoC
↓ Integrated Voice Product (validate → integrate → field → production)
From Technology to Your Product
Our engineering services validate and integrate our voice algorithms and AI models so they run on your real product data, your microphone and speaker structure, and your compute platform.
Rather than handing over algorithm or model files, we review the entire data flow — input capture, preprocessing, wake·location·speaker analysis, recognition, dialogue understanding, and product API execution.
We turn an algorithm's potential into your product's real performance.
mpWAV Technology (10 technologies)
↓ Engineering — Audio I/O · Data Processing · Model Optimization
Platform Porting · API Integration · Field Validation
↓ Customer Product
Service Portfolio
Confirm the effect on your real product data
We analyze single/multi-channel audio collected from your product and your current recognition results, comparing before and after mpWAV technology.
Covers
Connect the voice technology you need to your product
Apply a single module, or integrate the full software stack from preprocessing to recognition and dialogue.
Typical structures
Wake word, recognition, and dialogue inside the product
We review the feasibility of running mpWWD, mpASR, and mpLLM under your product's CPU·GPU·NPU and memory conditions.
Scope
Move validated technology to your target platform
We optimize technology proven on PC or reference environments to run in real time on your DSP, FPGA, or AP.
Scope
From voice input to the processing board
We connect the microphone array, speaker output, AEC reference, multi-channel audio I/O, and compute platform into one product structure.
Scope
Technology designed jointly around your product and data
When per-product optimization matters more than a standard module, we collaborate through algorithm licensing, AI model application, or co-development.
Scope
Dedicated voice interface structures for volume production
Building on FPGA·DSP validation, we review semiconductor IP or dedicated SoC collaboration for products that need power, size, and cost optimization.
Collaboration type
Official material presents voice interface chips and a dedicated SoC as business models and expansion directions. Today this proceeds as co-development and partnership, not a standard semiconductor product.
Discuss SoC·IP partnershipPoC | Software·SDK | On-Device AI | DSP·FPGA·AP Porting
Mic-Array·HW | Licensing·Co-development | Semiconductor IP·SoC
Proof of Concept
Voice technology performance depends on microphone count and layout, user distance, noise, speaker output, number of talkers, your existing ASR, and the execution platform.
mpWAV evaluates before-and-after results against data collected from your actual product rather than generic demo audio.
mpNC · mpAEC · mpBeamforming · mpAB
mpWWD · mpLocalization
mpS · mpDiarization
mpASR · mpLLM
Example outputs
Voice Enhancement: Noise·Echo·Distortion | Interaction: Wake Word·Localization·Speaker Segments
Recognition: WER·CER·Command Success | Dialogue: Intent·Slot·Context·API | System: Latency·Memory·Compute
Software and SDK Integration
Depending on your product's current structure and required features, apply a single module or integrate several technologies into one continuous pipeline.
Microphone → mpNC / mpAEC / mpBeamforming / mpAB → Existing ASR
Improves input quality without replacing your recognition engine.
mpWWD → mpLocalization → mpAB → mpASR → mpLLM → Robot Command
Connects wake-up, speech direction, noise/echo preprocessing, recognition, and dialogue.
Mic Array → mpAB → mpASR → mpLLM → Ordering API
Voice ordering and conversational option confirmation in store noise.
Single Mic → mpNC → mpWWD → mpASR → mpLLM → Device Function
For earbuds and small devices that can't add microphones.
mpAEC / Beamforming → mpS → mpDiarization → mpASR → mpLLM → Transcript / Summary
Connects multi-party processing, speaker separation, records, and summaries.
Existing ASR | Robot | Kiosk | Single-Mic Device | Meeting
On-Device AI Optimization
An on-device voice interface is not decided by model accuracy alone.
Wake-word detection, recognition, and the dialogue model must be reviewed as one execution structure against your product's memory, CPU·GPU·NPU, power, response time, and update method.
Wake-word detection
On-device speech recognition
Dialogue and intent understanding
mpLLM is being developed in the direction of a 1B-class on-device dialogue model; actual feasibility depends on model version, precision, memory, and the target accelerator.
Input preprocessing
Depending on your microphone conditions, the following applies as the front end:
Key review metrics
Audio Input → mpNC/mpAB → mpWWD → mpASR → mpLLM → Local Product API
Conditions: CPU · GPU · NPU · Memory · Power · Latency
Embedded Porting
Applying technology proven in a reference environment to a real product requires porting matched to the target processor's compute, memory, audio interfaces, and latency.
Official material presents FPGA implementation of the integrated mpAEC·mpBeamforming technology, MCU+DSP structures, AP·DSP porting, and multi-channel mic/echo input board design experience.
Multi-channel real-time processing
In-product low-latency processing
OS and product service integration
AI models and preprocessing together
mpWAV Reference Stack → Porting & Optimization
→ [ FPGA | DSP | AP | Edge AI ] → Audio Driver·ASR·API → Customer Product
Multi-Channel Hardware Integration
Applying multi-microphone technology to a real product means designing mic count and layout, channel synchronization, speaker output, the AEC reference, and the compute board together.
mpWAV brings experience with linear and hexagonal arrays, USB multi-channel audio I/O, FPGA·DSP·AP connections, and integrated speech preprocessing.
Design cases from official material
These are design cases from official material and do not imply common specifications for all currently supplied modules.
Mic Array + Speaker Output + AEC Reference → Multi-Channel Audio Board
→ FPGA·DSP·AP → mpAEC·Beamforming·Localization·mpAB → ASR·LLM·Product API
Licensing and Co-Development
Every product differs in microphones, users, noise, compute platform, and service goals — a standard package cannot solve everything.
mpWAV negotiates licensing, model optimization, or co-development scope based on your real data and product requirements.
mpNC · mpAEC · mpBeamforming · mpAB
Applied to your product's single- or multi-microphone input.
mpWWD · mpS · mpLocalization · mpDiarization
Connects wake word, user direction, multi-party voice, and speaker information to product functions.
mpASR · mpLLM
Reviewed for your domain vocabulary, menus, command sets, and product APIs.
Co-develops analysis models using per-product acoustic data — such as factory equipment anomaly detection.
Microphone arrays, audio I/O, preprocessing, recognition, dialogue, and product APIs in one project.
Technology proven in the first product extended to other mic configurations, platforms, product lines, and languages.
Product Requirements + Real-World Data → Joint Technology Review → Technology Stack Design
→ PoC → Product Optimization → Platform Integration → License·Production·Expansion
Semiconductor IP and SoC Partnership
Vehicles, smart devices, and volume products must weigh size, power, and cost alongside algorithm performance.
Building on software algorithms, real-time FPGA processing, and multi-channel hardware implementation, mpWAV reviews IP·SoC collaboration with semiconductor companies and volume customers.
Technology blocks under review
Fields under review
Collaboration phases
Algorithm Validation → FPGA Prototype → Architecture Review → Semiconductor IP → Dedicated SoC → Product Validation
Official material presents voice interface chips for vehicles, consumer devices, and defense, plus a dedicated SoC, as expansion directions. Public material does not include standard chip model names, datasheets, or production schedules.
Software → FPGA → Product Module → Semiconductor IP → Dedicated SoC → Mass Production
Targets: Performance · Latency · Power · Size · Cost
Service Packages by Application
Recommended technology
mpWWD · mpLocalization · mpAEC · mpBeamforming · mpAB · mpASR · mpLLM
Recommended services
Recommended technology
mpAB · mpASR · mpLLM · Mic-Array HW
Recommended services
Recommended technology
mpWWD · mpLocalization · mpAEC · mpBeamforming · mpAB · mpASR · mpLLM
Recommended services
Recommended technology
mpNC · mpWWD · mpASR · mpLLM
Recommended services
Recommended technology
mpS · mpDiarization · mpAEC · mpBeamforming · mpASR · mpLLM
Recommended services
Recommended technology
Noise-Robust Acoustic Processing · Anomaly Detection
Recommended services
Robot | Kiosk | Vehicle | Smart Device | Meeting | Factory
Find the Right Service
| Current situation | Recommended service |
|---|---|
| We want to confirm the effect of mpWAV technology first | PoC & validation |
| We want to connect a technology module to an existing product | Software·SDK integration |
| We want wake word, ASR, and LLM running inside the product | On-device AI optimization |
| Our target DSP·FPGA·AP is already fixed | Embedded porting |
| We need a microphone array and audio board | HW integration |
| We need per-product data and long-term collaboration | Licensing·co-development |
| We need a dedicated chip for volume production | SoC·IP partnership |
Confirm effect?→PoC | Connect to SW?→SDK Integration | On-device AI?→Model Optimization
Platform fixed?→Porting | Mic·board?→HW Integration | Long-term?→License/Co-dev | Volume?→SoC/IP
From Requirement to Deployment
The following is a typical example — the actual sequence and scope are adjusted per project.
Wake word, commands, dialogue, meeting records, listening improvement, or anomaly detection — we scope what you need.
Microphone count, speakers, user distance, number of talkers, dominant noise, and the target platform.
Single/multi-channel raw audio, the AEC reference, ground-truth text, and existing results.
Select the required technologies and processing order from the full mpWAV portfolio.
Confirm before/after performance and product features on PC or an evaluation environment.
SDK, model optimization, mic array, DSP·FPGA·AP porting, or licensing.
Connect audio input, mpWAV technology, your existing ASR, mpASR·mpLLM, and product APIs.
Tune latency, memory, compute, power, and model size.
Confirm results and user experience on the real product, where it's used.
Licensing, updates, follow-on products, international deployment, and SoC expansion.
Goals → System analysis → Data → Stack → PoC → Implementation → Integration → Optimization → Field → Production
Project Deliverables
The following are typical examples — actual scope is agreed per project under the development and licensing terms.
PoC Report | Software Library | AI Model | API | Ported Build | Hardware Module | Integration Guide
Implementation Experience
ASR preprocessing validated for home, showroom, and care robots
Multi-channel array I/O and ASR preprocessing structures deployed
Vehicle and mobility voice interfaces reviewed in real noise
Motor anomaly detection validated amid complex line noise
ClearSense Audio applied on smartphones and earphones, including welfare centers
Real-time FPGA, MCU+DSP structures, and AP·DSP porting
Robot voice test | Kiosk mic array | In-vehicle data collection
Factory motor measurement | ClearSense in use | FPGA·DSP boards
Project Information
You don't need every item confirmed to start — share the voice problems your product hits most often and the features you're aiming for.
FAQ
The full mpWAV portfolio can be reviewed.
That includes mpNC, mpAEC, mpBeamforming, mpAB, mpWWD, mpS, mpLocalization, mpDiarization, mpASR, and mpLLM.
Delivery status and contract form are agreed per project.
Yes.
Starting from your product structure and typical usage scenarios, we can design the required data channels, recording conditions, and ground-truth format first.
Yes.
mpNC, mpAEC, mpBeamforming, and mpAB apply in front of your existing ASR, improving the input audio.
mpNC covers single-microphone noise control feasibility.
Specific platforms and compute budgets are confirmed under real product conditions.
A microphone array, mpAB, mpASR, and mpLLM can be connected to interpret menu·options·quantity and hand off to the ordering API.
Real deployment also needs menu data, dialogue policy, and target hardware validation.
mpS and mpDiarization process multi-party audio and speaker segments, then connect to mpASR.
It is being developed in the direction of a 1B-class on-device dialogue model.
Feasibility depends on model version, quantization, memory, accelerators, and the target response time.
Official material presents real-time FPGA implementation, MCU+DSP structures, and AP·DSP porting experience.
Support for a specific chipset is reviewed against compute resources, the toolchain, and audio interfaces.
Multi-channel microphone arrays, speaker output, the AEC reference, USB audio, and processing board integration can be reviewed.
Whether it's a standard module or a custom design depends on the project.
There is no fixed standard duration or price.
They are agreed per project based on the technologies applied, data readiness, target platform, hardware scope, optimization level, and field validation scope.
The delivery form — source code, libraries, binaries, models, FPGA IP, or co-development — is decided by the contract terms.
Start Your AI Voice Project
Tell us your product type, mic·speaker structure, dominant noise, required features, and target platform — we will scope the right mpWAV technology and services.
From single-mic noise control to multi-channel echo and noise processing, wake·location·speaker analysis, on-device ASR and conversational LLM, DSP·FPGA·AP porting, and SoC expansion — start at whatever stage your product is in.
Real-World Audio → Technology Selection → PoC → Software/AI/HW Integration
→ Platform Optimization → Product Deployment (Earbuds·Robot·Kiosk·Vehicle·Meeting·Factory)
mpWAV develops and licenses voice interface technology that makes speech clear in noisy, real-world environments, built on 25+ years of speech signal processing research.
(주)엠피웨이브·사업자등록번호 859-81-01905·대표 박형민
Teilhard Hall 405, 35 Baekbeom-ro, Mapo-gu, Seoul, Korea
© 2026 (주)엠피웨이브. All Rights Reserved.
