<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN">
<HTML><HEAD>
<STYLE type=text/css> <!--@import url(E:\LMQ\LightingMail\\data\scrollbar.css); --></STYLE>

<META content="text/html; charset=utf-8" http-equiv=Content-Type>
<STYLE>BLOCKQUOTE{margin-Top: 0px; margin-Bottom: 0px; margin-Left: 2em}; </STYLE>

<META name=GENERATOR content="MSHTML 8.00.6001.19019"><BASE 
target=_blank></HEAD>
<BODY 
style="BORDER-RIGHT-WIDTH: 0px; MARGIN: 12px; BORDER-TOP-WIDTH: 0px; BORDER-BOTTOM-WIDTH: 0px; BORDER-LEFT-WIDTH: 0px" 
marginheight="0" marginwidth="0"><STATIONERY>
<DIV>
<DIV>&gt;&gt;&nbsp;There&nbsp;is&nbsp;also&nbsp;a&nbsp;IEEE&nbsp;paper,&nbsp;Adaptive&nbsp;Sampling&nbsp;Rate&nbsp;Correction&nbsp;for&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;Acoustic&nbsp;Echo&nbsp;Control&nbsp;in&nbsp;Voice-Over-IP,&nbsp;which&nbsp;introduced&nbsp;a&nbsp;complex&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;method&nbsp;to&nbsp;estimate&nbsp;the&nbsp;frequency&nbsp;offset&nbsp;and&nbsp;resynchronize&nbsp;the&nbsp;signals&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;using&nbsp;arbitrary&nbsp;sampling&nbsp;rate&nbsp;conversion.&nbsp;I&nbsp;wonder&nbsp;if&nbsp;it&nbsp;can&nbsp;provide&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;enough&nbsp;performance.&nbsp;Because&nbsp;I&nbsp;have&nbsp;also&nbsp;designed&nbsp;a&nbsp;sampling&nbsp;rate&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;converter.&nbsp;After&nbsp;tested&nbsp;the&nbsp;offset&nbsp;accurately,&nbsp;it&nbsp;can&nbsp;reduce&nbsp;the&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;offset&nbsp;to&nbsp;less&nbsp;than&nbsp;0.1Hz,&nbsp;then&nbsp;the&nbsp;signal&nbsp;after&nbsp;resampling&nbsp;is&nbsp;send&nbsp;to&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;speex&nbsp;AEC.&nbsp;But&nbsp;there&nbsp;is&nbsp;still&nbsp;hearable&nbsp;echo&nbsp;even&nbsp;if&nbsp;it&nbsp;is&nbsp;far&nbsp;less&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;than&nbsp;that&nbsp;can&nbsp;be&nbsp;heared&nbsp;before&nbsp;resampling.&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;Does&nbsp;anybody&nbsp;have&nbsp;any&nbsp;suggestion&nbsp;about&nbsp;practical&nbsp;acoustic&nbsp;echo&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;cancellation&nbsp;in&nbsp;low-cost&nbsp;soundcards?&nbsp;You&nbsp;know,&nbsp;most&nbsp;low-cost&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;soundcards&nbsp;have&nbsp;the&nbsp;problem&nbsp;of&nbsp;sampling&nbsp;rate&nbsp;asynchronous.&nbsp;</DIV>
<DIV>&gt;&gt;&nbsp;</DIV>
<DIV>&gt;&nbsp;That&nbsp;one&nbsp;sounds&nbsp;much&nbsp;more&nbsp;interesting.&nbsp;If&nbsp;you&nbsp;don't&nbsp;have&nbsp;access&nbsp;to&nbsp;IEEE&nbsp;</DIV>
<DIV>&gt;&nbsp;papers,&nbsp;you&nbsp;can&nbsp;find&nbsp;it&nbsp;at&nbsp;the&nbsp;author's&nbsp;university&nbsp;site.&nbsp;They&nbsp;don't&nbsp;do&nbsp;</DIV>
<DIV>&gt;&nbsp;anything&nbsp;extraordinary,&nbsp;but&nbsp;they&nbsp;have&nbsp;thought&nbsp;through&nbsp;how&nbsp;you&nbsp;can&nbsp;track&nbsp;</DIV>
<DIV>&gt;&nbsp;the&nbsp;sample&nbsp;rate&nbsp;offset&nbsp;by&nbsp;LMS,&nbsp;and&nbsp;use&nbsp;a&nbsp;continuously&nbsp;variable&nbsp;rate&nbsp;</DIV>
<DIV>&gt;&nbsp;converter&nbsp;to&nbsp;allow&nbsp;for&nbsp;it.&nbsp;</DIV>
<DIV>&nbsp;</DIV>
<DIV>Yes. As you said, this is still not a good solution.</DIV>
<DIV>&nbsp;</DIV>
<DIV>&gt; 
I&nbsp;noticed&nbsp;that&nbsp;the&nbsp;Fraunhofer&nbsp;Institute&nbsp;is&nbsp;now&nbsp;selling&nbsp;a&nbsp;package&nbsp;to&nbsp;&nbsp;</DIV>
<DIV>&gt; 
address&nbsp;echo&nbsp;cancellation&nbsp;when&nbsp;the&nbsp;tx&nbsp;sample&nbsp;rate&nbsp;cannot&nbsp;be&nbsp;trusted&nbsp;to&nbsp;&nbsp;</DIV>
<DIV>&gt; 
exactly&nbsp;match&nbsp;the&nbsp;rx&nbsp;sample&nbsp;rate&nbsp;-&nbsp;primarily&nbsp;in&nbsp;VoIP&nbsp;conferencing&nbsp;&nbsp;</DIV>
<DIV>&gt; 
applications.&nbsp;They&nbsp;say&nbsp;they&nbsp;use&nbsp;the&nbsp;spectral&nbsp;envelope,&nbsp;and&nbsp;disregard&nbsp;the&nbsp;&nbsp;</DIV>
<DIV>&gt; 
phase.&nbsp;That&nbsp;sounds&nbsp;like&nbsp;its&nbsp;not&nbsp;a&nbsp;million&nbsp;miles&nbsp;from&nbsp;the&nbsp;spectral&nbsp;&nbsp;</DIV>
<DIV>&gt; 
subtraction&nbsp;a&nbsp;lot&nbsp;of&nbsp;noise&nbsp;suppression&nbsp;schemes&nbsp;use,&nbsp;and&nbsp;those&nbsp;aren't&nbsp;&nbsp;</DIV>
<DIV>&gt; 
great&nbsp;at&nbsp;getting&nbsp;high&nbsp;levels&nbsp;of&nbsp;suppression.&nbsp;However,&nbsp;they&nbsp;claim&nbsp;very&nbsp;&nbsp;</DIV>
<DIV>&gt; 
high&nbsp;levels&nbsp;of&nbsp;echo&nbsp;suppression.&nbsp;There&nbsp;must&nbsp;be&nbsp;more&nbsp;to&nbsp;what&nbsp;they&nbsp;do&nbsp;than&nbsp;&nbsp;</DIV>
<DIV>&gt; the&nbsp;blurb&nbsp;indicates.&nbsp;</DIV>
<DIV>&nbsp;</DIV>
<DIV>&gt; 
It's&nbsp;relatively&nbsp;easy&nbsp;to&nbsp;getting&nbsp;high&nbsp;levels&nbsp;of&nbsp;echo&nbsp;suppression&nbsp;using&nbsp;&nbsp;&nbsp;</DIV>
<DIV>&gt; 
spectral&nbsp;subtraction&nbsp;methods.&nbsp;In&nbsp;fact&nbsp;even&nbsp;half-duplex&nbsp;cheap&nbsp;&nbsp;&nbsp;</DIV>
<DIV>&gt; 
hands-free&nbsp;phones&nbsp;can&nbsp;achieve&nbsp;that.&nbsp;The&nbsp;tricky&nbsp;part&nbsp;is&nbsp;not&nbsp;to&nbsp;distort&nbsp;&nbsp;&nbsp;</DIV>
<DIV>&gt; 
the&nbsp;"local"&nbsp;voice&nbsp;during&nbsp;double-talk.&nbsp;That's&nbsp;the&nbsp;hard&nbsp;part&nbsp;when&nbsp;you&nbsp;&nbsp;&nbsp;</DIV>
<DIV>&gt; can't&nbsp;rely&nbsp;on&nbsp;an&nbsp;adaptive&nbsp;filter.&nbsp;</DIV>
<DIV>&nbsp;</DIV>
<DIV>
<DIV>Does anybody know&nbsp;any kind of echo cancellation kernel which is not 
sensitive to different sampling rates?</DIV>
<DIV>At least I don't know.</DIV>
<DIV>But there&nbsp;is still a vivid example, AEC in MSN Messager, which is a 
real AEC, not a echo&nbsp;suppression.</DIV>
<DIV>It provides almost perfact echo cancellation even in double talk.</DIV>
<DIV>Why?</DIV>
<DIV>&nbsp;</DIV></DIV></DIV></STATIONERY></BODY></HTML>