Unsupervised overlap detection for single channel speech

This project is a python implementation of pyknogram extraction. Pyknograms are a modified verision of the STFT that emphasize more on the harmonic structure of speech. This work is used for co-channel speech analysis. Feel free to reach out if you need help to run this code.

Citations:

A detailed description of the algorithm is presented in our Transactions paper. Please cite:

[1] "Teager-Kaiser Energy Operators for Overlapped Speech Detection," Shokouhi, Hansen. IEEE/ACM Trans. on ASLP, 2017.

[2] "Robust Overlapped Speech Detection and its Application in Word-Count Estimation for Prof-Life-Log Data," Shokouhi, Ziaei, Sangwan, Hansen, ICASSP, 2015.

• Prereqs:
  • numpy
  • scipy
  • pylab

The directory "code" contains an example script (example.py) that calculates pyknograms for the wavefile selection.wav

Run the following inside the code directory:

$python example.py selection ../selection.wav

Outputs will be:

• figure1 : pyknogram
• figure2 : mean(spectral flux of pyknograms)
• figure3 : var(spectral flux)

Navid Shokouhi

December 2015