Voice Interactive Systems Technology Avionics (VISTA) Program.
Abstract
Although the preliminary test results are encouraging, it must be remembered that they were taken under ideal conditions. For all testing, the microphone was positioned just brushing the test subject's lips; however, a test was run with one test subject placing the microphone approximately four millimeters from the test subject's lips. The test results showed a 50% decrease in recognition accuracy for the same conditions as those with a microphone touching lips. Although the results are preliminary, it is apparent that the signal-to-noise ratio is a key factor in recognition accuracy. Another problem arises becuase of the automatic gain controls (AGC) found in most aircraft intercom systems. When there is no voicing for a period of time, the AGC increases the intercom sensitivity. If the first utterance spoken is intended for the recognizer it will likely be rejected because of the distortion caused by the AGC adjusting the gain during the utterance. This is demonstrated in the test results of all test subjects. No attempt was made to set the AGC before beginning the test; as a result, 90% of the first utterances were rejected which resulted in the lowering of the accuracy score by approximately 4%. The AGC has a release time of 10 seconds and the prompts are issued every second; therefore, after the first utterance the AGC has little effect. Some side tests were performed by making an utterance before signaling the computer to begin the test, and in each case the accuracy of the first test word increased to a point comparable to the other vocabulary words.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 18, 1982
- Accession Number
- ADA117288
Entities
People
- Lockwood W. Reed