 |
|
Speaker-dependent Speech Recognition and
Application to the Fabric Inspection System
|
| Sukeyasu KANNO, Katsuaki HAYASHI and Yuji YONEZAWA |
The performance of speech recognition degrades remarkably due to the non-stationary
noise and Lombard effect under heavy noisy environments. This paper describes
a practical method to recognize noisy Lombard speech , based on the detection
of voiced sound periods in factories. The effectiveness of three techniques,
i.e., WGD measure, word periods and weighted distances in voiced sound
periods for noisy Lombard speech, was confirmed through the word recognition
experiments using a 120-word vocabulary uttered by three male speakers.
Then, the approach was applied to the fabric inspection system using speaker-dependent
speech recognition. As experimental results, this system achieved high
recognition performance for a 61-word vocabulary and proved to be available
for an improvement of inspection efficiency in the factory.
|
|
| ■ |
Keywords ■■■ |
| |
speech recognition, Lombard effect,
voiced sound, fabric inspection system, noisy environment |
 |
|
PDFファイルをご覧になるためには、アクロバットリーダーが必要です。
ダウンロードはこちら |
|
