Products
 
Speech Recognition Engline  
Automatic Speech Recognition  
Embedded Speech Recognition  
Embedded Speech Synthesis  
Voice Tone  
Speaker Identification Engine  
Speech Morphing Solution  
   
 

Introduction to ThinkIT Embedded Speech Recognition Engine

Beijing Kexin Technology Limited

Introduction:

The ThinkIT Embedded Speech Recognition Engine (MSR) is designed especially for mobile equipments such as PDA, cellular phone, etc. The kernel is based on Hidden Markov Model (HMM), and the search algorithm has been optimized greatly for speed and accuracy suitable to the characteristics of embedded mobile devices. The Chinese acoustic model was obtained via training speech data with a variety of environments, speakers and accents. So the model should be robust to some extent. According to resources of different mobile equipments, the recognition engine can be customized with different configurations so that the overall performance is best to certain equipments.

Features and Requirements:

Speaker independent with accent tolerance

  Dynamic vocabulary configuration (deletion and insertion) with API functions

  Multilingual (Chinese and English) recognition

  Dynamic pronounce creation for new words

C interface API, easy for 3 rd party developing

  Running environment requirement :

CPU: 25MIPS plus DSP support for feature extraction

  Memory usage depend on vocabulary size (estimated with 200 words)

  RAM: 70KB (including dynamic memory usage)

  FLASH: 50KB

  ROM: 310KB

  Memory (RAM+FLASH) increase rate: 570 byte per word

Performance and Speed:

  Under normal usage environment (including in the car, on the street and in office), word accuracy is above 95% with vocabulary size of 200 words recognition accuracy is above 90% with the vocabulary size of 2000 words.

  Chinese full syllable accuracy: 1 best candidate: 80% ; 3 best candidates: 97% ; 5 best candidates: 99% .

  Real time speed if meets running environment requirement

Programming Models:

MSR engine is provided as a Windows DLL. It provides API which is a set of functions for you to call the engine. MSR API provides standard callback mechanism for applications to communicate with SR engine. Applications receive events through callback functions.

For more information about our speech products, please visit.

If you have any questions, please send email to support@thinkit.com.cn .

Disclaimer:

ThinkIt TM is a trademark of Beijing Kexin Technology Limited. All other trademarks are the property of their respective owners.

 
 
 
 
 
| HOME | COMPANY | PRODUCTS | CONTACT | COOPERATION |
Copyright for Beijing Kexin Tech Co. Ltd All Right Reserved