AMtrainer ---- A Praat script for quantifying the AM theory of intonation (Version 1.4.3) [Download]

by Yi Xu and Emily Lau

An interactive Praat script that allows you to:

_ _ _ _ _Original* _ _ _ _ _Resynthesis

* Original sound from ToBI training web site: www.ling.ohio-state.edu/research/phonetics/E_ToBI/. Original annotation + initial % and H, L after L*+!H

Explanation

AMtrainer offers a way to quantitatively test the AM theory of intonation (Pierrehumbert, 1980). It implements the AM assumption that intonation consists of tones that correspond to turning points in surface F0, and interpolations between them to fill the rest of the continuous F0 contours (Pierrehumbert, 1981).

The current version of AMtrainer also offers ways to vary the predictive power of the theory by implementing a number of schemes of abstracting the tones:

  1. No abstraction — Both the height and alignment of each tone are taken from the original sound, and surface F0 contours are generated by directly interpolating between adjacent tones
  2. Abstraction of pitch height — The categorical pitch value of each tone is obtained by averaging across all instances of its realization within the training folder
  3. Abstraction of tonal alignment as delay from syllable onset — The categorical alignment of each tone is obtained by averaging its real-time delay from the nearest preceding syllable onset across the training folder
  4. Abstraction of tonal alignment as relative location in a syllable — The categorical alignment of each tone is obtained by averaging its relative (i.e., proportional) location in the syllable where it is situated cross all its intances in the training folder

Instructions

  1. AMtrainer consists of only a single Praat script, which can be downloaded here.
  2. Put a copy of the script in the folder containing the sounds to be modeled;
  3. Launch the script by double-clicking it or opening it from within Praat;
  4. In the startup window, check or uncheck the boxes according to your need, and set appropriate values in the text fields or simply use the default values.
  5. Click OK and three windows will appear. The first window (PointProcess) displays the waveform together with vocal cycle marks (vertical lines) generated by Praat. This is where you can manually add the missing marks and delete redundant ones. You need to do this only for the named intervals, as explained next.
  6. The second window (TextGrid) displays the waveform and spectrogram of the current sound together with optional pitch and formant tracks in the spectrogram panel, and vocal pulse marks in the waveform panel. (These tracks and marks cannot be manually changed. So you can hide them to reduce processing time by using the corresponding menu.)
  7. At the bottom of this window are two TextGrid tiers. The top one is an interval tier for inserting interval boundaries and naming the intervals that you want to include in the modeling process. Tier 2 is a point tier for adding AM tones.
  8. You can make the script skip a voiceless region by making it a blank interval. Any blank interval with duration < minimum_pause_duration will be treated as a syllable-initial voiceless consonant.
  9. The third window (AMtrainer) is an interactive synthesis window. It displays the original f0 (blue dashed curve), synthetic f0 (red solid curve), transcript of the interval tier (top of the graph), and annotated tones (next to each tone). When there are no labeled intervals, only the original f0 is displayed.
  10. The AMtrainer window allows you to inspect the f0 contours in various ways: zooming in and out, scrolling left and right, and playing part or the whole of the original or resynthesized signal. The window also allows you to move to the next or previous sound in the folder.
  11. When you click "Next" or "Previous" in the AMtrainer window, the TextGrid and PointProcess windows will be refreshed, displaying the spectrogram, waveform and vocal cycle marks of the next or previous sound. You can repeat this process until all the sounds in the folder are processed. Or you can finish at any time by pressing "Exit".
  12. To implement any of the abstraction schemes, you need to select one of the abstraction methods from the dynamic menu:
  13. and then select "Get global tone parameters"
  14. After that, you can activate the synthesis window to check the synthesis results, with the dynamic menu set to the fitting schemes you want to test.

Output

Each time you press "Next" in the AMtrainer window, various analysis results are saved for the current sound as text files:

After going through all the individual sounds, global tone parameters can be obtained by running the script again and select "Get global tone parameters". The global parameters will be saved in AMmeans.txt, which will contain the same columns as in X.AMlabel_means for every individual sound file. The values of tone height, tone delay, tone location, however, will be averages of each tone category across the whole folder.

After executing "Get global tone parameters", tone parameters can be collected in ensemble file by running the script a gain and selecting "Get ensemble files". The ensemble tone parameters wil be saved in files with the following extensions:

  1. X.local_fitting — Containing f0 and time (relative to onset of sound) of all labeled tones
  2. X.time_delay — Containing f0 and time (relative to onset of each enclosing syllable) of all labeled tones
  3. X.relative_position — Containing f0 and time (location relative to duration of enclosing syllable) of all labeled tones
In addition, each of these files will contain two measurements of the model fitting accuracy:

How to cite

  • Lee, A., Xu, Y. and Prom-on, S. (2014). Modeling Japanese F0 contours using the PENTAtrainers and AMtrainer. TAL 2014. Nijmegen: 164-167.
  • Lau, E. and Xu, Y. (2019). Toward predictive modelling for AM theory of intonation. In Proceedings of The 19th International Congress of Phonetic Sciences, Melbourne, Australia.


    Yi's other tools