2006 Fiscal Year Final Research Report Summary
Establishment of Speech Communication under Very Heavy Environmental Noise
| Project/Area Number |
15500137
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Sensitivity informatics/Soft computing
|
|---|
| Research Institution | Yamaguchi University |
|---|
Principal Investigator |
UCHINO Eiji Yamaguchi University, Graduate School of Science and Engineering, Professor, 大学院理工学研究科, 教授 (30168710)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SUETAKE Noriaki Yamaguchi University, Graduate School of Science and Engineering, Associate Professor, 大学院理工学研究科, 助教授 (80334051)
AZETSU Tadahiro Yamaguchi Prefectural University, Faculty of Human Life Sciences, Lecturer, 生活科学部, 講師 (70285451)
|
|---|
| Project Period (FY) |
2003 – 2006
|
| Keywords | heavy noise / speech communication / bone conduction voice / air conduction voice / speech conversion / codebook / twin units SOM / self-organizing network |
|---|
| Research Abstract |
In general, a bone conduction microphone, which eliminates surrounding noise, is often used in extremely noisy environments such as engine rooms in ships or runways at airports. It detects the vibration of bones such as jaws, and it converts the vibration to voice. Unfortunately, the quality of this voice converted by this microphone is bad for a smooth communication. Therefore, the aim of this research is to develop an algorithm of voice conversion from a bone conduction voice to an air conduction voice, in order to supply a smooth communication method by voice in the extremely noisy environments. The results of this research are the following.
1. Voice Conversion by the Proposed TW-SOM
A new type of self-organizing map with twin units (TW-SOM), which can describe a nonlinear input-output relation with high accuracy, was proposed, and was applied to voice conversion. Concretely, TW-SOM learns a nonlinear relation between the bone and the air conduction voices by the twin units. After its learning, the bone conduction voice applied to TW-SOM is converted to the corresponding air conduction voice.
2. Verification of the Effectiveness of the Proposed Method and Application to Actual Ship-Handling Words
The effectiveness of the proposed voice conversion method was verified for actual ship-handling words by comparing with the conventional SOM and other competing neural network methods. It was also confirmed that the proposed method is more suitable for a hardware implementation than the other conventional methods.
3. Examination of Applicability to Other Fields
The key idea of the codebook used in the proposed method was successfully applied to an image expansion to get a clear image. TW-SOM is a general method to describe precisely the various nonlinear mappings including voice conversion. We would like then to examine its applicability to other fields as future studies.
|
