U.S. flag An official website of the United States government
  1. Home
  2. Inspections, Compliance, Enforcement, and Criminal Investigations
  3. Inspection References
  4. Inspection Guides
  5. Inspection Technical Guides
  6. Voice Recognition Systems
  1. Inspection Technical Guides

Voice Recognition Systems

[Previous Chapter] [Table of Contents]

 
DEPT. OF HEALTH, EDUCATION, AND WELFARE PUBLIC HEALTH SERVICE
FOOD AND DRUG ADMINISTRATION
*ORA/ORO/DEIO/IB*
Date: 11/10/87 Number: 52
Related Program Areas:
GLP, Devices

ITG SUBJECT: VOICE RECOGNITION SYSTEMS

A voice recognition system is an electronic computer controlled device which responds to the human voice. When hooked up to a computer terminal, it enables an operator to enter data merely by voice commands rather than having to manually enter the data using a keyboard.

You may encounter these voice recognition systems during your on-site inspections, most frequently where large amounts of data must be entered, generally for lists of numbers. These can be found during Good Laboratory Practice (GLP) inspections particularly for those aspects relating to laboratory analysis of various animal body fluids.

Typically, the operator will be wearing a microphone/headset while peering through a microscope counting various cells. With the voice recognition system, the numerical information can be entered into the computer merely by speaking the numbers aloud. The operator does not even have to lift his head away from the microscope.

Given the current state-of-the-art, the voice recognition systems are not universal, that is, they will not respond to just any voice. They have to be custom tailored to each operator who will be utilizing the system. In effect, the computer "memorizes" the way a particular person pronounces numbers and then will respond only to those numbers pronounced in that way.

Most programs have a training mode where the computer is "taught" to recognize that particular person's voice in pronouncing numbers. The software is written so that a numerical score can be granted to the recognition quality for that particular person. This is done by a validation procedure where the operator verifies the accuracy of the output compared to the input.

One feature normally found with the voice recognition systems is an audio repeat system where the synthesized computer voice repeats the word over a loudspeaker after the operator has enunciated it. In this way, the operators can keep their attention focused on the task at hand and yet make sure that the computer has "understood" the number that they were trying to enter. This aspect is quite important as generally no other positive verification is available.

During your on-site investigations when you find a firm utilizing a voice recognition system, you should try to observe the system while it is in use and determine the vocabulary limits for that particular system. Often only a certain limited range of numbers can be put into the system. You should also determine if such factors as speed of articulation, pitch, volume, ambient noise or other possible factors can corrupt data transcription. Operators may also be able to deactivate computer voice feedback as an irritant. There should also be a written SOP for both the computer "training" phase as well as the validation, tailored to each operator.

In the future, as the computer memory capacity expands, we may see additional use of these voice recognition systems.

[Previous Chapter] [Table of Contents]

 
Back to Top