Why D-Cheeps is
your best choice?
As the need for fast and accurate speech-to-text conversion grows across
various environments — such as 24/7 customer service, online lectures, and
meeting transcription — D-Cheeps delivers a high-quality, Korean-optimized
speech recognition AI that automates real-time speech-to-text conversion to
maximize operational efficiency.
D‑Cheeps is an STT software powered by the latest end-to-end (E2E) deep neural network
– based Korean speech recognition technology.
It supports real-time speech recognition as well as large-scale audio file processing
in various formats, and can be flexibly deployed to fit any customer environment.
BENEFITS
Benefit 01.
Automating Repetitive Tasks
By applying STT technology, tasks such as transcription, meeting documentation, and data entry can be automated. This reduces the time spent on manual work, allowing employees to focus on more strategic activities and improving overall operational efficiency.
Benefit 02.
Reducing Errors & Improving Accuracy
STT technology minimizes omissions and transcription errors that can occur during manual input, ensuring data consistency and accuracy. Users can rely on accurate results throughout every stage of data entry and processing.
Benefit 03.
Enhancing Productivity & Saving Time
STT technology instantly converts speech into text, accelerating information utilization and significantly reducing the time required for documentation. As a result, overall processing speed improves, boosting organizational productivity and decision-making efficiency.
Benefit 04.
Improving User Experience
A voice-command-based interface enhances accessibility and convenience, allowing users to interact with devices and services more intuitively. Even in hands-busy environments, voice input enables seamless operation, maximizing usability across diverse work settings.
FEATURE
powered by advanced speech recognition algorithms.
High-Quality, Reliable STT
-
Provides highly accurate speech recognition results
powered by deep learning and LLM-based algorithms -
Offers a stable, in-house developed solution with
flexible technology support tailored to customer needs
Scalable & Stable Operation
- Supports up to 50 full-duplex channels per server, with flexible scalability based on processing volume
- Operates on an L4 switch-based distributed
architecture for stable and reliable service delivery
Flexible Integration & Utilization
- Provides data APIs delivering word- and sentence-level confidence scores, timestamps, and metadata
-
Seamlessly integrates with add-on services such as
meeting minutes, captions, and call analysis
TECHNOLOGY
Real-time Speech Recognition
-
Real-time streaming and batch
speech-to-text conversion -
End-of-sentence (EOS) detection
metadata
Batch Speech Recognition
- Bulk transcription by uploading audio files
-
Supports multiple audio formats(WAV,
PCM, MP3) -
Supports multiple sampling rates
(8k, 16k, 44.1k, 48kHz)
Accuracy Enhancement
-
Hotword boosting and LM-assisted
inference -
Support for new vocabulary and
domain adaptation training
Speaker Diarization
-
Speaker-wise segmentation of input
audio
System Monitoring
- Monitor the STT system via a Health Check API
- Provide status per server component
Server Redundancy &
Scale-Up
- Active-Active redundancy configuration
- Docker Compose–based deployment,
scaling, and failover
User Interface & Settings
- Real-time result viewing/saving and
confidence analysis - Automatic extraction of the corresponding
audio waveform; recording settings - Configurable sampling rate and WAV file
saving options
Automatic Accuracy Calculation
- Recognition accuracy computed automatically
using Character Error Rate(CER), with result
export and analytics
ARCHITECTURE
API
API Parser
- Worker
Request - DB Query
- Engine
Manager
Request
Engine Manager
- Session Management
- User Management
- Database Management
DB
- Worker
Request - Post-processing Dictionary
- DB Query
- Recognition
Accuracy
Results - Engine Manager Request
- Server Status
Scheduler
- Job Queue Management
- Worker Mapping
Engine Worker
-
Worker 1
Worker N
D-Cheeps Library
- Voice Activity
Detection - Speech Recognition
- Speech Segment Detection
- Post-processing Algorithm
- Automatic Gain
Control - Hotword Boosting
- Feature Extraction
- Forced Alignment
- Config DB
Integration - Word Position
Tracking
USE CASES
Across domains where information is delivered via voice conversations or spoken commands,
D-Cheeps — our Korean-optimized speech recognition AI — powers voice-driven workflows.
Contact Center
AI Transcription
- Target Users
-
Customers of the National Police Agency’s call
center handling telecom and financial fraud reports - Service Overview
-
Automatically generates call transcripts by
converting voice phishing reports and civil
complaint calls into text in real time - Implementation Effects
-
-
Enables 24/7/365 automatic call recording and
database construction for all consultation
services -
Enhances citizen service quality and fraud
response capability through data-driven analysis
of consultation content
-
Enables 24/7/365 automatic call recording and
AI Secretary for
Aircraft Design Experts
- Target Users
- Engineers and staff of Korea Aerospace Industries
(KAI) - Service Overview
-
Allows AI assistants to receive and process voice or
text queries related to aircraft design in offices,
meeting rooms, and manufacturing sites - Implementation Effects
-
- Improves operational efficiency by providing
instant access to information anytime, anywhere - Enhances decision-making speed with 24-hour
AI-assisted service availability
- Improves operational efficiency by providing
Subtitles for
Broadcasts & Lectures
- Target Users
- Students enrolled in online university courses
- Service Overview
- Generates subtitles for lectures and broadcast
content in real time - Implementation Effects
-
- Provides accessible learning materials for not only
regular students but also hearing-impaired and
international learners - Improves learning effectiveness with real-time
subtitle support during lecture playback
- Provides accessible learning materials for not only