|
|
| |
|
|
|
|
|
|
 Click to enlarge image
|
 Origin |
Korea |
Products
- WinCE based Embedded SR SDK
- robust far-talking SR engine & DB
- Far-talking DSP module(SRB-200)
- Wall-pad embedding DSP module
- Kitchen TV phone,Wall-pad, other devices
- SR IR-Generator
- Trainable IR Generator for Appliances (Remote controller)
- Living room, Network with Wall-pad, SR DSP embedding
- SR Integrated Remote-controller
- Wall-pad, Appliances controller (IR training func.)
- Far-talking func. embeddin
Speech Recognition Switch KVS-1xA
- Set on existing switch control box without any wiring work
- Applicable to 1-4 bulb
- Far-talking SR in 6m
- Applicable to Home-network system
- Alarm, Sleep, Security mode Supply
- Time –telling function
- Speech command
- First, call SR switch name “Na-rae” when you start to control
"Hey Na-rae, illuminate living room"
"Hay Na-rae, what time is it?"
"Hay Na-rae, turn the light off in a moment"
Selling & Market
- Apply to Home-net &Construction Company
- OEM supply to Switch company
- Outlook replacing conventional Remocon Switch market
- Domestics : 500,000 EA needs per year
- Abroad : Eng/Chn/Jpn in 2009
Description
- 76 X 122 X 49(mm)
- Rated Voltage : AC 220V
- Rated Load :
P0 : 20w ~ 300w
P1 : 500w and below
P2 : 500w and below
VUI (Voice User Interface) Tech.
| |
Command&Control |
Large-vocabulary |
| Functions |
Menu control
Car control by on-air SR |
MP3 List/Movie List Recognition/Select
Address/Building (POI)search
Vocabulary Independent / Auto parsing SR
|
| Applied Format(Voice UI) |
Easy to speak
Don’t need to memory Commands
Keyword spotting
Speaker Adaptation support |
2Stage Based fast large-vocabulary Recognizer
-1 Stage : Fast search
-2 Stage : Detail search
Real-time list parsing and network building using G2P
Address search : Tree structure
-1stage : province (Seoul, Jeju….)
-2 stage : street (N-best output)
-3 stage : Continuous Number recognition
Building(POI)search Tree structure
-1 stage : province
-2 stage : POI search (N-best output) |
| System environment |
CPU : 533MHz
Memory : 4MB
OS : WinCE, QNX, Linux |
CPU : 533MHz
Memory : 15MB
OS : WinCE, QNX, Linux |
| Speed |
1 sec |
7 sec (million words) |
| Robustness |
80Km-Driving, Window closed, Audio Off
-Recognition Rate : 95%
80Km-Driving, Window open, Audio On
-Recognition Rate : 90% |
80Km-Driving, Window closed, Audio Off
-Recognition Rate : 90%
80Km-Driving, Window open, Audio On
-Recognition Rate : 85% |
| Test conditions |
Simulation with TEST DB
Driving Test |
Simulation with TEST DB
Driving Test |
Competitors & Tech. Comparison
| |
KoreaPowerVoice |
Conventional SR(ETRI, MS, etc.) |
| Units |
Continuous Keyword Spotting |
Isolated word recognition |
| Form |
On-air : no need any Hot-key for SR
Real-time continuous SR |
Remocon or Handle-Key needed
Recog after end-point detection |
| Robustness |
High recognition rates in noisy environments by noise cancel tech.
High recognition rates in driving with window-open |
Recognition rate drops in noisy environments
Impossible to recog in driving with window-open |
| Conveniences |
No need to see the monitor
No need to hear the beep to record |
Speaker must see the monitor
Speaker must hear the beep to record |
| Hot-Key |
No Need |
Needed (H/W, high cost) |
Nuance(US)
- VoCon3200 – Large - vocabularyspeechrecognitionengine
- Weaktonoise – hardtofindVADinnoisyenvironments
- Majority : VariouslanguageDBready
KoreaPowerVoice(KOR)
- trigger-word based continuous speech recognition engine
- Robust to noise –95% recognition rate in 100dB noise (Driving, MP3, Window-open)
- On-air Speech recognition by OOV technique
- Easy to port Embedded system (DSP, WinCE) by managing resources
|
|
|
|
| |
|
|
|
Related Keywords:
Bookmark / Share this page