Nautilus Systems, Inc. logo and menu bar Site Index Home
News Books
Button Bar Menu- Choices also at bottom of page About Nautilus Services Partners Case Studies Contact Us
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [Subscribe]

Re: DM: Speech Enhancement -- Using Neural Nets?


From: Mark Last
Date: Wed, 20 Oct 1999 21:20:47 -0400 (EDT)
Hi Anjani,

Check the following paper:
A.L. Gorin, S.E. Levinson, A.N. Gertner and E. Goldman, Adaptive
Acquisition of Language, Computer Speech and Language, vol. 5, no. 2, 
pp.
101-132, Apr. 1991.

They are using a neural-like structure, called "information-theoretic
connectionist network", which is based on Shannon's information 
theory.

Mark

At 08:41 PM 10/19/99 MDT, anjani avasarala wrote:
>Hello,
>
>       can anyone help me in tacking the below problem?.
>       I am working on speech enhancement.
>        We are trying to address the issue of speech enhancement
>        when the speech is degraded by two ways:
>        1. Additive noise
>        2. This case needs a brief introduction:
>           The situation is like this. The speech signal is
>           encoded using a standard coding technique (say
>           Continuously variable slope delta modulation (CVSD)).
>           So all we have is now a binary bit pattern. If this
>           is all, there is no problem. We can apply corresponding
>           decoding technique to get back the signal. But now some
>           bits in this coded speech are flipped (the choice of bits
>           is at random). So, the sequence information is lost.
>           In such cases we would like to do the following studies:
>           (a) Is there any way to reconstruct the speech signal from
>           this new binary pattern ( i.e., flipped coded speech)?
>           (b) What features of the original speech signal does this
>           new signal retain?
>
>         This problem can be looked in two different ways:
>         1. Applying bit error correction codes. But this
>            methods doesn't take into account the speech concepts.
>         2. Another method is exploiting the redundancy in the
>            speech signal. As we know speech is highly redundant
>            in nature. So, it is not necessary to know each and
>            every word to infer the message. So, does coded speech
>            still contain this redundancy?. If so,  what happens when

>            bits are flipped at random?.
>            Can neural networks be useful in this context?.
>            Currently, I am working on neural network methods only.
>
>           Hope I am clear. Write to  me for further details.
>           Hope someone will respond to this.
>          Thankyou,
>Anjani
>
>____________________________________________________________________
>Get free email and a permanent address at 
>http://www.netaddress.com/?N=1
>
>
---------------------------------------------
Mark Last
Visiting Assistant Professor
Computer Science and Engineering
University of South Florida
4202 E. Fowler Ave., ENB 118
Tampa, FL 33620, USA
Tel: 813/974-4763
Fax: 813/974-5456
mlast@csee.usf.edu
WWW: http://www.csee.usf.edu/~mlast/




[ Home | About Nautilus | Case Studies | Partners | Contact Nautilus ]
[ Subscribe to Lists | Recommended Books ]

logo Copyright © 1999 Nautilus Systems, Inc. All Rights Reserved.
Email: nautilus-info@nautilus-systems.com
Mail converted by MHonArc 2.2.0