Following the tuning process in the tutorial, I am trying to optimize the threshold for keyword search. I met some problems, however, in the plotting of ROC curve. To determine detection rate is no problem, but I don't know how to calculate the false alarm rate. I know how to count the number of flase alarm, yet, don't know how to define the denominator of flase alarm rate. Could anybody give me some advices?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
According to the tuning process in the tutorial, in the 1 hour long recording of movie sound with few occurrences of my keyword (only 1 keyword), I know how to count the occurence of flase alarm but I don't know how to define the denominator of false alarm rate. Is it the total number of keyword search in the 1 hour period? If it is, how to get the total number of keyword search in the 1 hour period?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Denominator is a number of keyword detections. You run pocketsphinx_continuous, it will output you 5 detected keywords, some of them are true, some false. This 5 is the denominator.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Following the tuning process in the tutorial, I am trying to optimize the threshold for keyword search. I met some problems, however, in the plotting of ROC curve. To determine detection rate is no problem, but I don't know how to calculate the false alarm rate. I know how to count the number of flase alarm, yet, don't know how to define the denominator of flase alarm rate. Could anybody give me some advices?
Denominator is the amount of signalled keywords (both false and true).
According to the tuning process in the tutorial, in the 1 hour long recording of movie sound with few occurrences of my keyword (only 1 keyword), I know how to count the occurence of flase alarm but I don't know how to define the denominator of false alarm rate. Is it the total number of keyword search in the 1 hour period? If it is, how to get the total number of keyword search in the 1 hour period?
Denominator is a number of keyword detections. You run pocketsphinx_continuous, it will output you 5 detected keywords, some of them are true, some false. This 5 is the denominator.