Acoustic Analysis with Python and ESPS

Kyle Gorman
Department of Linguistics
Institute for Research in Cognitive Science
University of Pennsylvania

kgorman@ling.upenn.edu

Under construction...

Using esps.py Get esps.py Acknowledgements
Python ESPS Source Linguistics IRCS University of Pennsylvania

A lot of people like using Praat. Some people even like to use Praat with a Python or Lisp scripting engine. Most everything I want to do, though, is implemented by ESPS and a bit of simple analysis. esps.py is a single well-documented Python module that reads and writes Praat TextGrids to and from Python lists, does F0 analyses, calculates signal intensity (RMS and intensities across different frequency bands), extracts spectral slices. and does LPC formant analysis.

Prerequisites

esps.py depends on both Python and external software.

Several modules depend on NumPy, a free numerical computing package for Python.
The normalization, squashing, and fitting operations depend on stats.py, which in turn depends on pstat.py (available here and here, both mirrored here). Make sure you are using the recent versions of both and not the ones that depend on Numeric.
Lastly, an installation of ESPS is a must (source code has been open-sourced and is available for free here), except for the two TextGrid functions. None of the graphical (Xwaves) components are used, so an installation of X11 is not necessary. The software in this package depends on the following programs from ESPS.
- get_f0
- sgram
- formant
- hditerm
- pplain

Get `esps.py`

All the modules are contained in a single file, esps.py. This is a change from earlier versions, where different types of analyses had different modules; this now allows me to more easily maintain the software. You can obtain esps.py here. In a change from earlier versions, all documentation is now 'inline' in the module as block comments. Please refer to this.

Acknowledgements

Catherine Lai and Stephen Isard both were extremely helpful in compiling all kinds of software; also, in discussing polynomial fitting and ESPS, respectively. Thanks also to Joshua Tauberer regarding binary search.

[ ~ ]