<HTML><HEAD>
<META http-equiv=Content-Type content="text/html; charset=iso-8859-1">
<META content="MSHTML 6.00.6000.16481" name=GENERATOR></HEAD>
<BODY style="MARGIN: 4px 4px 1px; FONT: 10pt Tahoma">wow, the attitude wasn't necessary. What I was asking wasn't in the documentation, how am I supposed to know I wasn't supposed to initialize the encoder/decoder every time (I was trying to save space for the allocs), I've never dealt with codecs before and it isn't in the documentation. I was not insinuating that your code was wrong, I was hoping you could tell me that I set it up improperly, which you did. you didn't have to be a dick about it.<BR><BR>>>> Jean-Marc Valin <jean-marc.valin@usherbrooke.ca> 08/03/07 9:13 AM >>><BR>
<DIV style="COLOR: #000000">> on the decode side I encoded a voice on a windows machine. I took the<BR>> encoded information and transferred it over to the C54x project and<BR>> decoded it and it didn't sound that great. I watched the decode process<BR>> and it would appear that the first 40 of the 160 words (2 bytes) are all<BR>> 0. <BR><BR>Speex (just like any other speech codec) is stateful and this is the<BR>(*one time*) latency you're seeing. Nothing unusual.<BR><BR>> this means that every 20ms there is 5ms of silence which is what I<BR>> believe is causing our poor decode quality or at the very least an<BR>> indicator that something isn't right. <BR><BR>Yes, it's your code that isn't right. If you're seeing those 40 zeros<BR>every time, it means you're re-initialising the codec for every frame<BR>(or something along those lines).<BR><BR>> the submode is set to 3 so I<BR>> don't think it thinks it's in another mode. I know that the decode on<BR>> the windows machine doesn't do this because the output file for the<BR>> windows machine is continuous data after the first 160 0's. it seems to<BR>> me like it decodes part of it, but then doesn't decode the rest. if you<BR>> would like the encoded file and the output please let me know.<BR><BR>You're not seeing that because you're probably using speexdec on Windows<BR>(which strips those samples away) and comparing to software you wrote on<BR>the C54x.<BR><BR>> on the encode side I encoded a few seconds of voice using my C54x<BR>> platform and tried to decode it on windows. the resulting decode<BR>> produced a voice that sounded much like the C54x decoded voice that was<BR>> encoded by windows (bad). I will provide the encoded data to anyone who<BR>> wants it. I don't notice any pattern of 0's in the decode or anything<BR>> like that. I want to say that it's the sample rate that is causing the<BR>> problems but when the G729A ran at this rate it worked quite well.<BR><BR>No, it's your code that is causing the problem.<BR><BR>> if there is anything that you can do to help I would really appreciate<BR>> it.<BR><BR>Yes, fix it.<BR><BR> Jean-Marc<BR></DIV></BODY></HTML>