A Praat plugin for fast(ish), accurate, (nearly) automatic formant-tracking. [wiki] [YouTube]
Fast Track is a Praat plugin that makes accurate formant tracking fast and easy. Fast Track automatically runs multiple formant analyses on a given sound file, and tries to pick the best one for you (surrounded by extra boxes in the example below).
Fast Track has built-in functions to help along every step of a vowel analysis project from start to finish:
-
Extract vowel sounds from larger recordings with TextGrids files.
-
Automatic selection of the 'best' analysis.
-
Automatically generate images of analysis for data validation. All alternate Analyses are saved for easy corrections.
-
Quick workflow for manually editing formant tracks.
-
Aggregate fine-sampled data into coarser measures (e.g., average formants for every 20% of duration).
-
Automatic generation of images showing final analyses for data validation.
-
Automatic generation of data files containing fine-sampled acoustic analysis calculated every 2 ms along the vowel.
Comparison of multiple analyses:
Image of winning analysis:
CSV file containing analysis information (formant frequencies and bandwidths, f0, harmonicity, intensity, etc.) sampled every 2 ms from the start to the end of the sound:
Output
The plugin can analyze to a single file or to an entire folder of sounds at once. Fast Track generates (among other things):
-
CSV files containing the frequencies of F1-F4, predicted formant values (for error checking), formant bandwidths (F1-F4), f0, intensity, and harmonicity, each sampled every 2 ms (by default).
-
Images for the visual verification of final analyses, and for the comparison of alternate analysis.
-
A log of the selected analyses, which can then be used to select alternate preferred analysis.
-
Detailed information regarding the analyses carried out for each sound.
-
A new (as of 6/23) 'lite' version that doesn't save any intermediate files.
Algorithm
1) Formant tracking is carried out at multiple maximum-formant settings, always looking for 5.5 formants.
2) The goodness of each candidate is established by considering the smoothness of the formant trajectories using a regression analysis (and optional heuristics).
3) The 'best' analysis is selected as the winner. Images are made for visual verification.
4) The user accepts the automatic selection, or indicates an alternate analysis.
5) The user is given an opportunity to manually edit winners to ensure completely-accurate tracks.
For more information please see the wiki! Or check this paper out.
If you use this, please cite it:
Barreda, S. (2021). Fast Track: fast (nearly) automatic formant-tracking using Praat. Linguistics Vanguard, 7(1). https://doi.org/10.1515/lingvan-2020-0051
Its the only way I can tell if people use it and what they use it for. Thanks!
Author
Santiago Barreda - UC Davis, Linguistics (https://linguistics.ucdavis.edu/people/sbarreda)
License
This project is licensed under the MIT License - see the License file for details
Acknowledgments
This work is the implementation of the ideas presented in several previous works:
- Nearey, T. M., Assmann, P. F., & Hillenbrand, J. M. (2002). Evaluation of a strategy for automatic formant tracking. The Journal of the Acoustical Society of America, 112(5), 2323-2323.
- Wassink, A. B., & Koops, C. (2013). Quantifying and interpreting vowel formant trajectory information. In Best Practices in Sociophonetics Workshop at NWAV (Vol. 42).
- Weenink, D. (2015). Improved formant frequency measurements of short segments. In ICPhS.
- Zhang, C., Morrison, G. S., Enzinger, E., and Ochoa, F. (2013). Effects of telephone transmission on the performance of formant-trajectory-based forensic voice comparison – Female voices. Speech Communication, 55(6), 796–813. https://doi.org/10.1016/j.specom.2013.01.011