ECE 590SIP, Spring 2018, is an informal seminar on automatic spectrogram reading. Sample code for writing and displaying spectrograms will be provided (possibly in Praat, possibly in ipython, maybe a bit of both). Sample waveforms will also be provided, and some puzzles.


Slides and Code

I-python notebooks will be posted here:

  1. Jan 17: Wideband and Narrowband Spectrograms (jupyter source)
  2. Jan 31: Manner (jupyter source)
  3. Feb 7: Vowels (jupyter source)
  4. Feb 14: Semivowels (jupyter source)
    lecture video
  5. Feb 28: The Crossword-Puzzle Method for Decoding
    • Slides containing a mystery spectrogram: PPTX, PDF
    • The same spectrogram, with IPA guesses and word guesses that were generated during our group spectrogram reading: PPTX, PDF
    • HTML and IPYNB showing how the spectrograms were computed.
    • video
  6. Mar 7: Fricatives. HTML and IPYNB
  7. Mar 14: Nasals. Yupik. HTML and IPYNB
  8. Mar 28: Stops and Affricates. Bengali HTML and IPYNB
  9. Multilingual Data
  10. Software and References
  11. How to Start a Praat Transcription
    • Method #1: From Scratch
      1. Read the waveform in Praat
      2. Select the waveform, and choose Annotate - To TextGrid (silences)...
      3. Select both the waveform and the new TextGrid, choose View and Edit, and start adding Intervals
    • Method #2: From Unaligned Transcription
      1. At the command line, type pip install praatio --upgrade
      2. Download uniformTextGrid.py
      3. Create a transcriptions.txt file with your wavfiles and transcriptions, one per line. Phones should be separated by spaces, words may be separated by sil.
      4. Type python uniformTextGrid.py transcriptions.txt outputPath
      5. Load the waveform, and its associated new TextGrid, into Praat