CMSC 25030: Computational Models of Speech
Spring Quarter 2008


Announcements

  • Matlab tips from class are posted for your reference below.

Coursework Deadlines

  • Tuesday, May 6: Final Module 1 Report Due
    • 4 single spaced pages, plus figures.
    • One report for each group.

Contact Information

  • Instructor: Partha Niyogi
    Office: Ryerson 167
    Email: niyogi at cs
    Website: people.cs.uchicago.edu/~niyogi/
    Office Hours: By Appt.

  • TA: Aren Jansen
    Office: Ryerson 257b
    Email: aren at cs
    Office Hours: By Appt. (for now)

Suggested Reading

  • Discrete-Time Signal Processing by Oppenheimer, Schafer and Buck

    This is a good reference for Fourier analysis.

  • Statistical Methods for Speech Recognition by Jelinek

    This is thorough presentation of HMM-based speech recognition methods.

Course Materials

  • Matlab Tips and Tricks: These are the notes from class on Tuesday, April 29. Also, see the link at the end for a huge number of other examples.
  • timitread.m: This function takes a TIMIT wav filename and returns an array containing the waveform.
  • praat.tar.gz: This archive contains a directory 'praat' (make sure this is placed in your home directory) with the following contents:
    • praat: Praat binary compiled for the CS department Linux machines.
    • transcribe.praat: A Praat script to automate the speech transcription process.
    • trans: The target directory for transcriptions.
    • .praat-user-startUp: Praat initialization script to create a transcribe button for sound objects (i.e. runs the transcribe script). This file must be copied to your home directory to function.

Interesting Links

  • Interactive Vocal Tract Demo

    This useful demo allows you to vary the articulatory configuration while displaying the resulting phonemes produced.

  • Sinewave Speech Examples (British English)

    Here are some examples of sinewave speech along with the original recording. If you listen to the sinewave version first, it will be very difficult to to understand. However, after you've listened to the original, the sinewave speech becomes miraculously intelligible. You can also download the Praat script so you can generate sinewave versions of your own recording. (see Praat software link below)

  • Praat Speech Analysis Software

    This is a powerful speech analysis package. Playing with it can be a lot of fun.

  • Speech Sound Generator Software

    This program performs a simulation of the vocal tract for producing vowel sounds. It allows you to adjust the vocal tract configuration real time to see the affect on the sounds produced. This is a simple example of articulatory synthesis.

  • AT&T Natural Voices Text to Speech

    This is a state-of-the-art concatenative text-to-speech (TTS) demo. Concatenative TTS involves stringing clips of speech together to simulate words. The voice "Crystal" is the best developed.