How to analyze a folder

Select the "Fast Track > Track folder..." button. You will be asked to indicate your working directory. This is where Fast Track will find the sound files to analyze, and place all output files.  

Place all files you want to analyze in a subfolder called 'sounds' in the working directory. Each file should be cut so that it contains only sections meant to be tracked. This is because sections without formants will cause unpredictable errors when trying to determine the best analysis. Please see preparing sounds for more information.

 

Depending on the number of sound files, number of analysis steps, and computer speed, this process can potentially take a long(ish) time (≈30-45 minutes for 1000 files at 20 steps (analyses) per sound). Also, it can't be interrupted other than by closing Praat (see bugs page).

The process is divided into three steps to be outlined below. Each steps can be run (or re-run) independently, assuming the previous step has already been run. The default is to run all three steps at once, however, there is no benefit to doing this. If you have a particularly large amount of data to be analyzed, it may make sense to run the three steps independently even the first time.

0. Settings

One thing to consider is that if the folder contains speech from a single person, setting an appropriate (and narrow) range of analysis frequencies greatly increases the probability of finding the correct analysis. In my experience, most errors lead from using cutoff frequencies that are very wrong for the voice being analyzed.

If your folder contains speech from many people with substantially different formant ranges (e.g., adult males and children), you may get worse performance for everyone! This is because performance really benefits when the analysis ranges are reasonable for the data being analyzed.

Please see the section on 'Getting a good analysis' for more information.

1. Track formants

This step creates all the candidate analyses and saves them all as formant objects. Outputs are:

• formants: a folder containing Praat formant objects representing all candidate analyses for each sound. The naming convention is filename_N, where N is the analysis step number.

• infos: a folder containing text files with information about the analyses being carried out.

• fileList.Strings: a Praat Strings object containing the list of audio files analyzed is also saved.

2. Autoselect winners

This step automatically selects the best track from the candidates. The default smoothness metric is the sum of the median absolute deviation across the formants being analyzed. Outputs are:

• images_comparison: a folder containing images comparing the different candidate analyses. Winners are marked with extra boxes. Analyses are arranged in images from top to bottom and left to right in terms of increasing analysis frequency.

• winners.csv: a CSV file indicating the winning analysis frequency for each sound. Analysis steps are numbered in terms of increasing analysis frequency.

• csvs: a folder containing CSV files. Each file contains information about formants, f0 and intensity for each sound.

• More information is added to the info files in the 'infos' subfolder.

3. Get winners

In the final step, the best analyses are collected. Final analyses are determined based on the contents of the 'winners.csv' file. Outputs are:

• formants_winners: a folder containing copies of the winning formant objects. The naming convention is 'filename_winner_', where 'filename' is the sound file name.

• csvs: the CSV files in the 'csv' folder are updated.

• images_winners: a folder containing images of the winning analysis for each sound.

• processed_data: the data in the CSV files is aggregated into a single file, and the regression coefficients are collected into another file, and both are placed in this folder.

4. Fixing Mistakes (optional)

If you want to override the automatically-selected analysis, you can do this in one of two ways:

1. If some alternate analysis is better, update the value in the column marked 'winner' and the columns for F1-F4. If no good analysis exists, write the number 0 in the winner column and the file will not be collected or aggregated.

2. If no single analysis is very good, a different analysis can be specified for each formant by putting a different number in each column. For example, analysis 10 could be used for F1, F3 and F4, while analysis 15 could be used for F2.

After changing the winners in the winners.csv file, the third step (get winners) must be run again.

You can also manually edit formant tracks. This can be done by placing copies of the winning formant objects you want to edit in the 'formants_final' subfolder. Make sure they are copies because the winning formant files will be deleted (deletion would not be a big deal as these can be renegerated easily and quickly by repeating step 3 above).

After placing copies of winning formant files into the 'formants_final' subfolder, select the "Fast Track > Edit tracks (folder)..." button. This will loop through each file in the folder. The user has the option of accepting the track, re-tracking, or manually editing. Once the track is accepted, a final copy is saved ('filename_final'), and the 'filename_winner_' version is erased from the subfolder. Optionally, CSV files and images can also be generated at this point.