INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING Introducing the INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING located at Mississippi State University Department of Electrical and Computer Engineering Box 9571, Mississippi State, Mississippi 39762 Tel: 601-325-3149 Fax: 601-325-3149 email: picone@isip.msstate.edu MISSION STATEMENT Mississippi State University for over 100 years has had a mission of being a center of excellence in the State of Mississippi for: • Learning — to enhance the intellectual development of its students • Research — to extend the present limits of knowledge • Service — to apply its research to improve the lives of people The Institute for Signal and Information Processing (ISIP) offers a multidisciplinary program focused on the development of next generation information processing techniques. Research at ISIP is centered on intelligent information processing, perhaps the most important technology of the next century. ISIP draws upon a wide range of research experience in areas such as signal processing, communications, natural language, database query, intelligent systems, and discrete controls. Its present vision is to develop systems capable of intelligent interactions with users by the integration of a multiplicity of interface technologies including speech, natural language, database query, and imaging. S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 1 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING SIGNAL PROCESSING RESEARCH AT MISSISSIPPI STATE UNIVERSITY IS MULTIDISCIPLINARY Anthony Skjellum Robert J. Moorhead High Performance Computing Image Processing Computer Science / ERC ERC Victor A. Rudis Stephen E. Saddow Forestry Imaging Semiconductor Technology USFS EMRL Communications Laboratory Bud Rizer Elect. and Comp. Eng. Assistive Technologies T.K. Martin Center Joe Picone Signal Processing Inst. for Signal and Info. Proc. S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 2 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING Outside World (hub #0): • Allied Telesyn MR 820T Domain: isip.msstate.edu • 10BaseT 8 port hub (10 Mbits/sec) • Cat-5 Unshielded Twisted Pair • 155 Mbits/sec ATM (campus) isip00 (fileserver, router, and domain server): • Sun SPARC 5 • 70 MHz MicroSPARC II • 32 Mbytes RAM, 1 Gbyte local disk • 2 ethernets (for routing) • 60 Gbytes magnetic disk (Seagate Elite) Exabyte 10h Tape Library • 8 mm tapes • 70 Gbyte capacity • 140 Gbytes compressed Sharp JX-325 Color Scanner: • one-pass 24-bit color scan • 300 dpi native mode isip01 (compute server): • Sun SPARC 20-512 • Two 50 MHz SuperSPARC Processors • 192 Mbytes RAM, 1 Gbyte local disk isip02 (demo machine): • Sun Sparc 5 • 70 MHz MicroSPARC II • 32 Mbytes RAM, 1 Gbyte local disk • T1 Telecom Interface datlink 0 and datlink 1 (audio): • Townshend DAT-Link+ • 16-bit digital audio • AES/EBU and SP-DIF isip03 and isip05 (compute server): • dual Pentium Pro • 200 MHz Processor • 256 Mbytes RAM, 1Gbyte local disk isip04 and isip06 (laptops): • Samsung Sens 810, Toshiba Tecra 500 CDT • 133 MHz Pentium Processor • 40 Mbytes RAM, 2 Gbyte local disk ncd20c00 (clients): • NCD Xterms S I • 16-bit audio S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 3 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING ISIP’S FOCAL PROJECT • An Integrated Services Transactions Processor That Supports Advanced Telecommunications Interfaces such as an Asynchronous Transfer Mode (ATM) Digital Communications Link Example: Telephone-Based Natural Language Query of Entertainment Archives Customer : “Give me all movies, uh, make that only the recent movies, directed by Martin Scorsese and starring Robert DeNiro, and oh, by the way, make that movies about gangsters only.” Computer : We have three titles available (the titles of the movies are shown on the television screen with real-time video of promo clips from each movie below the title). Please select a movie. Customer : “That one with the three guys looks good, I’ll take that one. I want it to start at 8:00 PM tomorrow.” Computer : (The promo clip for the selected movie starts playing on the television.) The movie titled GoodFellas starring Robert DeNiro and directed by Martin Scorsese will be delivered for viewing on your television on Thursday, September 25 starting at 8:00 PM. Thank you for using ISIP’s Entertainment Server. Good-bye. Local Central Office ATM (160 Mbps) Unix Multiprocessor (Sparcstation 2000): • Voice • 8 Processors • Video • 512 Mbytes of memory • Data (X Windows) • videotape jukebox S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 4 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING A T1-BASED DATA COLLECTION SYSTEM FOR SUN/UNIX WORKSTATIONS S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 5 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING Speech Recognition “Show me all the reports from Language Text the White House on Healthcare.” Model Language Model Natural Language Tagged Text Processing Semi-Parser Flat Parsed Natural Structures Language Understanding Knowledge Extractor Knowledge Filled Templates Extraction Request Generator Netscape Requests Netscape S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 6 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING PARALLEL IMPLEMENTATIONS OF FAST FOURIER TRANSFORMS • Object-oriented software implemented in C++ COMPUTER ANALYZER CLASS UTIL UTIL CLASS TRANSFORM INT_4 FLOAT_8 ANALYZE COMPUTE_STATS COMPUTE • Detailed performance analysis in a common framework FFT ORDER Algorithm 16 64 256 1024 4096 16384 RAD2 20 60 280 1960 10900 97100 RAD4 20 60 250 1800 9720 58220 SRFFT 20 40 160 1060 6140 38100 FHT 20 40 140 640 3800 38100 QFT 20 40 160 880 6560 44020 DITF 20 60 360 2500 12320 104080 (Table entries are computation times in usec) S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 7 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING BASIC TECHNOLOGY: A PATTERN RECOGNITION PARADIGM BASED ON HIDDEN MARKOV MODELS ∏ i ( , , … ) Recognized Symbols: P S O ( ) = arg max P W t ( O t O t ) – 1 T i i Language Model: P W t ( ) i i Prediction P O t W t ( ) P W t ( ) i O t Search Algorithms: P W t ( ) = - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - P O t ( ) i P O t O t i [ , , , … W t ] Pattern Matching: W t ( ) – 1 ( , , ) Signal Model: P O t ( W t W t W t ) – 1 + 1 S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 8 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING THE JEIDA JAPANESE COMMON SPEECH DATA CORPUS . Number of speakers 150 speakers 75 male speakers 75 female speakers Number of items per speaker 323 items monosyllables 178 isolated words 35 4-digit sequences Number of repetitions per item 4 repetitions of each item Range of speaker age 20 yrs. to 60 yrs. Amount of data 120 hours Number of Digital Audio Tapes 76 (120-minute tapes) Total number of utterances 193,800 utterances Number of channels/mic. type 2 (dynamic and condenser mics.) Anticipated size of final corpus 6.5 Gbytes (16-bit 16 kHz samples (13 CD-ROMs uncompressed) @ 1.0 secs per utterance) S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 9 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING AUTOMATIC GENERATION OF N-BEST PROPER NOUN PRONUNCIATIONS NEURAL NETWORK SOLUTION E P S T AI _ N OUTPUT PHONEME 100100111001000 OUTPUT REPRESENTATION OUTPUT LAYER • • • • • • • • • • • • HIDDEN LAYERS • • • • • • • • • • • • • • • • • • INPUT LAYER 10100000001000011101101001 INPUT REPRESENTATION E P S T E I N CONTEXT LETTER WINDOW S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 10 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING JAVA APPLETS http://isip.msstate.edu/software/java_system_response Other ISIP Java Applets include: • Convolution • Frequency Response • Nyquist Criterion • Analog and Digital Filter Design • Compilers and Assembly Code • Hidden Markov Model Toolkit • Speech Recognition Primer S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 11 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING SYLLABLE-BASED SPEECH RECOGNITION FOR CONVERSATIONAL TELEPHONE SPEECH S I S I I I P P s h h s p p c c ee ee
OCTOBER 2, 1997 TELECOMMUNICATIONS PAGE 12 OF 16 INSTITUTE FOR SIGNAL AND INFORMATION PROCESSING ECHO CANCELLATION FOR SPEECH RECOGNITION HYBRID IN NETWORK ANNOUNCER OR ECHO ( ) SPEAKER, s n ( ) a n A CALLER, (CUE FOR SPEAKER ID) AUTOMATIC SPEECH ( ) ) s n ECHO RECOGNIZER FOR CANCELLER SPEAKER IDENTIFICATION S I S I I I P P s h h s p p c c ee ee
Recommend
More recommend
