Date: March 4, 2005

Time: 3:00 pm

Location: GCATT Room 325

Speaker(s): Dr. Tomohiro Nakatani, Visiting Scholar, NTT Corp. (Japan)

Title: Blind Dereverberation of Single Channel Speech Signals Based on Harmonicity

Abstract:
Long reverberation is a serious problem that degrades distant speech applications in the real world such as hands-free speech recognition. One simple way to overcome this problem is to dereverberate the signals prior to the recognition, but this is still a challenging problem especially when using a single microphone. In this talk, I will propose a new dereverberation principle based on an inherent property of speech signals, namely harmonicity. This principle allows us to estimate the inverse filter of a room transfer function by obtaining a filter that enhances the harmonicity of observed speech signals. We developed a dereverberation method by extending this principle, which can provide a good approximation of the inverse filter when a sufficiently large amount of utterances are available. Simulation experiments show that this method can achieve high quality speech dereverberation in terms of reverberation energy decay curves and automatic speech recognition (ASR) even when the reverberation time is 1.0 sec.

In addition, I will give a brief introduction of speech research activities at NTT Communication Science Labs, which includes an integrated system of blind source separation (BSS), ASR, and question answering (QA).

Bio:
Tomohiro Nakatani received his BE, ME, and PhD from Kyoto University, Japan in 1989, 1991, and 2002, respectively. He joined NTT Corporation as a researcher in 1991, and conducted research on speech enhancement technologies for intelligent human-machine interfaces with Dr. Okuno and Dr. Kawabata. From 1998 to 2001, he belonged to business departments of NTT and NTT-east Corporations, and promoted multimedia services. Then, he returned to a research section of NTT Corporation in 2001. Since 2005, he has been visiting Prof. Juang's research group in School of ECE, Georgia Institute of Technology. Currently, he is a senior research scientist of NTT Communication Science Labs, NTT Corporation and a visiting scholar of Georgia Institute of Technology. He received the 1997 JSAI Conference Best Paper Award and the 2002 ASJ Poster Award. He is a member of IEEE, ASJ, IEICE, and JSAI.

Slides: sem03_04_05_Tomohiro_Nakatani.pdf

Date:	March 4, 2005
Time:	3:00 pm
Location:	GCATT Room 325
Speaker(s):	Dr. Tomohiro Nakatani, Visiting Scholar, NTT Corp. (Japan)
Title:	Blind Dereverberation of Single Channel Speech Signals Based on Harmonicity
Abstract:	Long reverberation is a serious problem that degrades distant speech applications in the real world such as hands-free speech recognition. One simple way to overcome this problem is to dereverberate the signals prior to the recognition, but this is still a challenging problem especially when using a single microphone. In this talk, I will propose a new dereverberation principle based on an inherent property of speech signals, namely harmonicity. This principle allows us to estimate the inverse filter of a room transfer function by obtaining a filter that enhances the harmonicity of observed speech signals. We developed a dereverberation method by extending this principle, which can provide a good approximation of the inverse filter when a sufficiently large amount of utterances are available. Simulation experiments show that this method can achieve high quality speech dereverberation in terms of reverberation energy decay curves and automatic speech recognition (ASR) even when the reverberation time is 1.0 sec. In addition, I will give a brief introduction of speech research activities at NTT Communication Science Labs, which includes an integrated system of blind source separation (BSS), ASR, and question answering (QA).
Bio:	Tomohiro Nakatani received his BE, ME, and PhD from Kyoto University, Japan in 1989, 1991, and 2002, respectively. He joined NTT Corporation as a researcher in 1991, and conducted research on speech enhancement technologies for intelligent human-machine interfaces with Dr. Okuno and Dr. Kawabata. From 1998 to 2001, he belonged to business departments of NTT and NTT-east Corporations, and promoted multimedia services. Then, he returned to a research section of NTT Corporation in 2001. Since 2005, he has been visiting Prof. Juang's research group in School of ECE, Georgia Institute of Technology. Currently, he is a senior research scientist of NTT Communication Science Labs, NTT Corporation and a visiting scholar of Georgia Institute of Technology. He received the 1997 JSAI Conference Best Paper Award and the 2002 ASJ Poster Award. He is a member of IEEE, ASJ, IEICE, and JSAI.
Slides:	sem03_04_05_Tomohiro_Nakatani.pdf