Welcome
Welcome and thank you for visiting my website. I am Takuya Yoshioka, working on speech technologies at NTT Communication Science Laboratories. I received a PhD degree of Informatics from Kyoto University in 2010. I am a member of IEEE and ASJ.
Send me your comments and questions at yoshioka.takuya(at)lab.ntt.co.jp. Your feedback is very appreciated.
Research Summary
I have been studied speech enhancement in reverberant environments. My research has covered the following topics. Click on the titles for more detail.
- Dereverberation
The quality of speech signals recorded in hands-free situations is degraded by room reverberation. Dereverberation is a technology for automatically cancelling the reverberation effect to recover the speech quality. The dereverberation technology may be applicable to hearing aids, sound editors, teleconferencing systems, and automatic speech recognition systems. It is also useful for other kinds of speech processing technologies including direction of arrival estimation, beamforming, and blind source separation, whose effectiveness is unduly limited in reverberant environments. Furthermore, when applied to music signals, the dereverberation technology provides a way of controlling the reverberation effect. I developed several dereverberation methods with my collegues. - Joint noise suppression and dereverberation
In typical hands-free situations, speech signals are contaminated by not only room reverberation but also ambient noise. Thus, dereverberation needs to be performed in conjunction with noise suppression. I developed two methods that effectively combine dereverberation and noise suppression techniques. - Joint blind source separation and dereverberation
Blind source separation (BSS) is a technology for separating mixtures of multiple sounds. This technology has been studied extensively and significant progresses have been made over the last decade. However, conventional BSS methods perform poorly when the reverberation time is large. Several researchers have addressed this problem, but it is still an open question. In order to overcome this limitation, I developed a method for unifying BSS and dereverberation, which thereby realizes BSS even under highly reverberant environments.
Biography
| Apr. 2006 - Present | Researcher at NTT Communication Science Laboratories, Japan |
| Mar. 2010 | PhD degree of Informatics from Kyoto University, Japan |
| Mar. 2006 | Master's degree of Informatics from Kyoto University, Japan |
| Jun. 2005 - Mar. 2006 | Trainee at NTT Communication Science Laboratories, Japan |
| Aug. 2004 | Summer Intern at NTT Communication Science Laboratories, Japan |
| Mar. 2004 | Bachelor's degree of Engineering from Kyoto University, Japan |
Contact Information
Tel: +81 774 93 5326
Fax: +81 774 93 5158
E-mail: takuya(at)cslab.kecl.ntt.co.jp
2-4, Hikaridai, Seika-cho, “Keihanna Science City”Kyoto 619-0237, Japan