嶄猟井

 

 

About MSIIP
 
   
   
Introduction  
History
News
 
 
 
 

 

   

About MSIIP -- History

 

 

 

  In 1986, led by Prof. Zuoying Wang, several departments of Tsinghua University, including Department of Radio Electronics(renamed to Department of Electronic Engineering in 1987), Department of Automation, Department of Precision Instruments, Department of Foreign Language, etc., jointly applied for and undertook the project "Computer Natural Language Translation" belonging to the first batch projects of "863" Program of China, which consisted of three sub-projects: speech recognition, machine translation and speech synthesis. This project carried out in 1987. Prof. Wang was responsible for the overall project coordination and research on the sub-project "Computer Speech Recognition". Meanwhile, Prof. Wang organized research team and established speech recognition technology laboratory(SRTL) of Department of Electronic Engineering Tsinghua University.

  After the establishment of SRTL, substantial pioneering work was carried out on Chinese speech recognition. In May of Year 1987, Professor Wang proposed Duration Distribution Based Hidden Markov Model(DDBHMM), which was able to describe the real distribution of speech units instead of imposing exponential distribution of segment length on speech units according to the classic Hidden Markov Model(HMM). Different from the classic HMM, there was post-effectiveness using DDBHMM. Subsequently Professor Wang proposed training algorithm and recognition algorithm for DDBHMM, succeeding in the establishment of the DDBHMM overall solution and forming a more comprehensive speech recognition theory framework.

  SRTL was not only a pioneer on speech recognition research, but also a frontier in industrial system design and development. In 1988, SRTL was the first to achieve  speaker-dependent whole-syllable Chinese isolated-word speech recognition system based on Dynamic Time Warping(DTW). In 1992, isolated Chinese character speech recognition system, THED919 Dictation Machine, jointly developed by SRTL and China National Electronic Devices Corp., passed the identification of Ministry of Electronic.

  In addition to the steady research sponsorship from national "863" Program, the project also brought a powerful stimulation to the entire speech recognition industry in China. Chinese speech recognition technology had been rapidly developing with the hardwork of numerous domestic research institution and researcher, including SRTL. In year 1994,1995 and 1998, "863" intelligent computer system experts continuously organized evaluation of speech recognition technology for three times, the content of which extended from speaker-dependent isolated-word recognition to continuous speech recognition and finally reached speaker-independent continuous speech recognition. SRTL won three consecutive champions, establishing the domestic leading position on speech recognition technology.

  In order to convert advanced technology into real-life products, based on the continuous speech recognition system for the evaluation of "863" program in 1998, SRTL launched THEESP system in 1999, the first domestic online real-time Chinese continuous speech recognition system with full independent intellectual property rights. Speech Processing Department of Koninklijke Philips N.V. ever purchased the right to use the system and took advantage of it to develop the Chinese version of the continuous speech recognition product "FreeSpeech".

  From year 2002, considering the application prospect of speech recognition technology, SRTL strated R&D on IVR(Interactive Voice Response) platform oriented multi-channel concurrent speech recognition engine and limitative resources oriented embedded speech recognition engine. During the same year, Beijing Tianlang-tech Co., Ltd. was initiated by Tsinghua University and Beijing High Technology Venture Capital Co., Ltd., to promote the industrial development of speech recognition technology. In Year 2003, Tianlang launched the first domestic telephone speech recognition engine supporting multi-channel concurrence. In the same year, "VoiceCards", an embedded speech recognition product for Symbian platform, won the prize for excellence of Individual Efficiency Class of Nokia-Symbian S60 Application Contest. Although Tianlang almost stopped operation in 2004 due to various reasons, as one domestic pioneer in industrialization of speech recognition technology, it has made the contribution to the development of speech recognition industry.

  In 2006, Tsinghua-iFlyTek Joint Laboratory for Speech Technologies was founded for the strategic alliance with the domestic leading enterprise on speech industry, Anhui USTC iFLYTEK Co., Ltd. While persisting in the research on multi-platform and multi-application oriented speech recognition technology, SRTL began to enter the field of speech search technology. Later, SRTL undertook the project "Exploration Research on Content-based High Performance Speech Search Technology" belonging to the "863" Program of the Eleventh Five-Year Plan for National Economic and Social Development, and had successively completed technology prototypes and demo systems for navigation-oriented speech search system for massive place-names(Voice Search for POI), massive speech documents oriented content retrieval system(Spoken Term Detection, STD), audio matching and filtering system, etc.

  Along with the extension of research interests and contents, the lab formally renamed to "Multimedia Signal and Intelligent Information Processing Laboratory"(MSIIP) in 2009. Aimed at contents implied in multimedia signal, MSIIP has been carrying out the research on the cognition, understanding, efficient management and utilization of data and information, adopting massive data oriented processing technologies including statistical pattern recognition, machine learning, data mining, etc.

 

 
     
 
 

Tsinghua University  |  School of Information Science and Technology  |  Department of Electronic Engineering  |  USTC iFLYTEK

 
 

Copyright@Multimedia Signal and Intelligent Information Processing Laboratory