1997 Fiscal Year Final Research Report Summary
The Study of Voice Control System to Increase Tractor Safety
| Project/Area Number |
07660339
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
農業機械学
|
|---|
| Research Institution | Mie University |
|---|
Principal Investigator |
SATO Kunio Mie University, Bioresources, Associate Professor, 生物資源学部, 助教授 (20126973)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
HOKI Makoto Mie University, Bioresources, Professor, 生物資源学部, 教授 (00024589)
|
|---|
| Project Period (FY) |
1995 – 1997
|
| Keywords | tractor / safety / voice recognition / control / vowel / cepstrum / neural network / spectrum analysis |
|---|
| Research Abstract |
Following results have been obtained.
(1)Because vowel sounds in the human voice have continuous and identifiable structures in the quefrency domain, voice recognition technology could be applied to tractor control even in the presence of large amounts of engine noise.
(2)A control system for the electric components of a tractor was developed to test a vowel recognition system using the outline of the vowel pitch. The electric components considered, consist of the head lights and turn signal lights. Varying engine idle speeds with the tractor at rest, an average vowel recognition rate of 99.1% was achieved. While moving at a constant engine speed of 1600 rpm a recognition rate of 90% was achieved.
(3)A word recognition system, programd to a specific operator, performed at a 100% recognition level.The same system performed at 95% recognition for other operators.
(4)An evaluating function for the vowel period was developed successfully to recognize the unusual length of vowels that are characteristic of intentional or unintentional utterances for emergency stops.
(5)A microphone-pair was used and a method was developed to distinguish vowels uttered by the principle operator from other voices outside the tractor. The ratio of pitch-sound-pressures between the two microphones can be used to identify the operators voice successfully.
(6)Further investigations are needed to combine individual research results. However, the fundamental voice recognition methods using vowel pitches in the presence of large scale engine noise has achieved the status of word recognition. The future of this research will be to combine individual technologies to establish a practical application for current market vehicles. The goal is to raise level from word recognition to natural language recognition for tractor control system.
|
Research Products
(9 results)
