Speech Processing
Joe Campbell

Johns Hopkins University
Whiting School of Engineering
525.747

Speech Processing Class Information

• Class Schedule and Grading
• Join the Mailing List
• Exam Coverage
• Syllabus
• Speech Processing 525.747 Course Description, Prerequisites, etc.
• Speech Processing, JHU PTE WSE Catalog
• Windows 95B & Windows NT 4.0 SP3 Dial-Up Networking (DUN) JHU APL/MCC login script

Course Materials

• Homework Q&A, project Q&A, and exam tips (class digest)
• Class Text & Software
  • Deller, Hansen, and Proakis. Discrete-Time Processing of Speech Signals (our textbook)
  • Matlab (preferred programming environment for this class)
    • Matlab (commercial)
    • Matlab Student Version
    • Mastering MATLAB (supplemental text)
    • Free Alternatives to Matlab
      • octave
      • RLaB
      • Scilab
• Speech Processing References
• Russian /a/ and /i/ synthetic phonemes. 10 tube approximations of Fant's data. Matlab binary "mat" file format:
  • ReadMe file for Algorithm 2 samples
  • /a/ Fs=10000 F0=70 P=40 K=55
  • /a/ Fs=8000 F0=70 P=40 K=55
  • /i/ Fs=10000 F0=70 P=30 K=60
  • /i/ Fs=8000 F0=70 P=30 K=60
  • All the above in a Zip archive

Laboratory

• Animated Convolution (more to come)
• Audio Morphing, etc.
• DSP First: A Multimedia Approach (excerpt) with Spectrograms, etc. (and Intro to Discrete Time Systems)
• Interactive Data Previewer w/Matlab!
• Some Interesting Sounds (and Auditory Demonstrations CD)
• Speech Synthesis and Text-to-Speech Conversion
  • AT&T Demos
  • Synthetic Speech Examples
  • Talking Web Page
• Speech Coding for Wireless Communications
• Speech, etc. Coding for the 'net
• Listen to Speech Coders (LPC, CELP, MELP, etc.)
  • Speech Compression Demo (LPC, CELP, VSELP, G.711, G.723, G.726, G.728, G.722, etc.)
  • Speech Coding Demo (ADPCM, LD-CELP, CELP, and LPC) (alternate)
  • Audio demonstration
  • The importance of perceptual weighting and unimportance of SNR in 4800 bps CELP coding - let your ears be the judge (each file is 361 kB):
    • CELP with perceptual weighting (FS-1016): SNR = xxx dB
    • CELP without perceptual weighting: SNR = yyy dB
• MELP (new federal standard algorithm for 2400bps voice communications and Federal Information Processing Standards Publication "FIPS PUB")
  • The U.S. Department of Defense Digital Voice Processor Consortium (DDVPC)
    • Hear MELP, CELP, LPC, and CVSD
    • Download MELP 1.2 source code, alternate download
  • ARCON, the DDVPC's Test Director
  • MELP Draft Standard
  • Vendor demos and software: ASPI and TI's C3x Demos, Tools, and MSP50C30 Long-duration speech processor and 8-hour recorder
• Audio Tools (PC, Mac, and Sun audio players, etc.)

Reference Materials

• Speech Processing References
• The comp.speech USENET news group Frequently Asked Questions and answers (FAQ) (alternate)
• Lots of speech software
• Speech Research Links (UCSC)
• International Telecommunication Union (ITU) Standards (purchase G.7xx recommendations & software)
• European Telecommunications Standards Institute (ETSI) (free GSM standards & software)
• Audio Signal Processing in Humans and Machines (Berkeley)
• Speech Coding and Compression
  • mu-law coding (alternate)
  • Download CELP and LPC software and audio files
  • FS-1016 CELP Implementation Tips & Download CELP 3.3
  • DDVPC's FS-1016 CELP
  • FS-1016 CELP and FIPS Pub 137 LPC-10e (comp.speech FAQ) (alternate)
  • FS-1016 CELP on the Motorola DSP56001 (comp.speech FAQ) (alternate)
  • Request LPC-10e FS-1015/FIPS Pub 137 Matlab code (Painter & Spanias)
  • Request CELP, GSM, etc. Matlab code (see Painter & Spanias, A MATLAB Software Tool for the Introduction of Speech Coding Fundamentals in a DSP Course, ICASSP-96, p. 1133)
• Speaker Recognition
  • Example Speaker Recognition Systems (see "Voice")
  • Speaker Recognition (comp.speech Q6.6) (alternate)
  • Speaker (and Speech) Recognition (comp.speech links) (alternate)

Related Topics & Sites

• JHU: Center for Language and Speech Processing
• MSU: Vincent Voice Library
• OGI: Center for Spoken Language Understanding
• PSU: Acoustics Dept
• Internet Phones
• Khoros - multidimensional (e.g., image) processing. This package began at the University of New Mexico. It features objected oriented block signal processing. Note that many users prefer version 1.05 over the newer version 2.02.
• Khoros Speech Toolbox
• Free Compilers
• Hidden Markov model software (Matlab anyone?)
  • Joe Picone's discrete HMM - C++ source code
  • Richard Myers' C++ source code (old version with web page)
  • Jialong He's discrete and continuous HMMs - GNU C source code
  • Speech Training and Recognition Unified Tool (STRUT)
• TI's new family of 'C6x DSP chips are here and here along with lots of software
• Acoustical Society of America
• IEEE Signal Processing Society