Data Format

  • The ROC HR folder contains folders of keywords that you would like Hit Rate performed upon.
  • Each of those folders must have pairs of .wav files and .csv files that each have the same name except for the extension.
    • Each pairs .wav file should contain multiple utterances of a single unique keyword.
  • The utterances should be spaced out such that the beginning of the next utterance’s rising edge comes no quicker than one second after the falling edge of the previous utterance.
  • The .csv file in the pair details the start and stop times for the keyword utterances in the .wav file.
  • The first start time is always 0, and the first end time is right before the rising edge of the next utterance.
  • Each start time thereafter is always equal to the previous end time, and each end time follows the same rule of right before the rising edge of the next utterance.
  • Each row must also have the keyword AS IT WAS STATED IN THE DAK INPUT. This means that if there is a class called “Okay”, but it is spelled as “ok” in the DAK input and labeled “okay” in the .csv file, the ISDK will not work as expected and may even throw an error/crash.
  • This nomenclature must also be followed for naming the folders that hold the file pairs. Below is a small example to display these concepts: