|
|
Back Number > > Topics > > 2005 > > October
|
|
|
| |
| The information on this website is presented "as is." Product availability, organization, and other content may differ from the time the information was originally posted. Changes may take place without notice. |
| High-performance, real-time sound separation device developed |
|
October 24, 2005
Kobe Steel, Ltd. has developed technology to separate specific sounds from
ambient background noise. This is the first time in the world that sound
separation can take place on a real-time basis with a high degree of accuracy.
Along with the technology, Kobe Steel has successful developed the world's
smallest sound separation device, which fits in the palm of a hand. The
sound separation program is imbedded in a digital signal processor (DSP),
making possible the small size.
Humans are able to concentrate on one sound among a mixture of sounds, for
example honing in on a person speaking at a noisy party. However, microphones
cannot make a distinction and hear all sounds at the same intensity.
Kobe Steel hopes the new technology can eventually be used across a spectrum
of applications where it is necessary to clearly isolate specific sounds.
Examples where clear voice recognition could be applicable include voice-controlled
car navigation systems and cell phones used in noisy settings. Diagnosing
the operating sound of machinery for timely maintenance and repair offers
possibilities for industrial use. In the near future, the sound separation
system may also be helpful to humanoid robots that need to understand human
voice commands.
Kobe Steel's device separates audio signals from a variety of sound sources.
With a minimum of two microphones, the device uses algorithms to analyze
and separate the audio signals into independent sound sources. Using speakers
or headphones connected to the sound separation unit, it is possible to
filter out the background noise and hear only the targeted sound.
Algorithms for sound separation have already been developed by other researchers
and have been unveiled at a number of academic conferences. However, they
require an enormous amount of processing power and processing time. Even
processors used in PCs do not have adequate capabilities to separate mixed
sounds in the real world. Until now, real-time processing tended to result
in low sound separation quality. If accuracy was stressed, processing time
became lengthy because of the complicated calculations involved.
To overcome these problems, Kobe Steel and the Nara Institute of Science
and Technology jointly developed their own sound separation algorithm that
can analyze sound source signals and separate independent sound sources.
Using a high-performance signal processor, the device is able to quickly
crunch data. Excellent performance under real conditions and sound separation
on a real-time basis were achieved.
Kobe Steel's sound separation device uses off-the-shelf microphones and
a standard high-performance processor to keep costs relatively low. The
device runs on batteries and is considered to be the world's smallest of
its type.
This sound separation research was conducted in collaboration with Associate
Professor Hiroshi Saruwatari of the Graduate School of Information Science
at the Nara Institute of Science and Technology, a national university in
Nara, noted for its work in speech recognition and sound source separation.
Kobe Steel was responsible for the practical application of the algorithm
and evaluating the technology under real conditions.
 |
|
|
Back to Top
