<div>Hi JM,</div> <div> </div> <div>Thanks for the earlier replies. I have tested it with quality 1,2&3 the result is not much different. Also I must tell you that I am using the armbe-linux-gcc compiler which does not suppoort arm5e-asm optimisations. I get lots of assember related errors. Is there any alternative? Will moving on to a board supported by arm-linux-gcc compiler ,to enable arm5e optimisation, help?</div> <div> </div> <div>Nav<BR><BR><B><I>Jean-Marc Valin <jean-marc.valin@usherbrooke.ca></I></B> wrote:</div> <BLOCKQUOTE class=replbq style="PADDING-LEFT: 5px; MARGIN-LEFT: 5px; BORDER-LEFT: #1010ff 2px solid">Quality 0 has not been entirely converted to fixed-point, so it actually<BR>runs much slower on ARM. It's very poor quality anyway, so I don't see<BR>why you'd want to use it.<BR><BR>Jean-Marc<BR><BR>Naveed Rehman a écrit :<BR>> Hi JM,<BR>> <BR>> I am using narrowband mode with speex qaulity set to 0 and
complexity<BR>> set to 4. The code works fine on the 3.2GHz processor. The test<BR>> program speexenc takes 7ms to encode 5sec worth of voice on the<BR>> 3.2Ghz processor. The same takes 700ms on the ARM5E intel Xscale. The<BR>> code that I wrote puts 10 speex frames in one RTP packet. When<BR>> exexcuted on ARM I get lots of underrruns hence noise. I was<BR>> wondering perhaps its because of the speed of ARM processor as<BR>> dictated by the significantly slower speexenc on ARM. I hav'nt gone<BR>> much into speex but I guessed that under 20ms to encode 20ms of voice<BR>> (in speex packet) would get rid of underruns and hence noise. Am I<BR>> going wrong somewhere..<BR>> <BR>> TIA nav<BR>> <BR>> Jean-Marc Valin <JEAN-MARC.VALIN@USHERBROOKE.CA>wrote:<BR>>> Can any body please tell me the compiler options to optimise the <BR>>> speex code for ARM TE architecture. Fixed point is enabled. It <BR>>> currently takes
about 700ms to encode 5 sec worth of voice. I need<BR>>> to get it down to under 10ms.<BR>> <BR>> The ARM port has suffered a bit from the fact that I haven't been<BR>> able to keep the assembly up-to-date. Don't know what<BR>> bit-rate/options you're using so it's hard to say much about the<BR>> absolute numbers. The only thing I can say is that 10ms for 5 seconds<BR>> of speech would mean using only 0.2% of your 533 MHz chip, or about 1<BR>> MIPS. That's simply not possible using any modern speech codec, so<BR>> you'll have to either change that requirement or deal with<BR>> uncompressed (or very lightly compressed) audio.<BR>> <BR>>> Is the option --enable-arm5te-asm supported? I am using speex<BR>>> 1.1.12 version. The configure.ac file does not seem to support <BR>>> --enable-arm5te-asm option. I tried --enable-arm5e-asm but the<BR>>> result is'nt much different. Real time is not possible at that<BR>>>
rate. Processor is Xscale 533MHz.<BR>> <BR>> --enable-arm5e-asm should be fine, but yes it needs improvements. <BR>> Basically, the only thing interesting (from the POV of Speex) in the <BR>> ARM5E arch is the DSP-like instructions that can do 16x16 multiplies<BR>> in a single cycle.<BR>> <BR>> <BR>> --------------------------------- Try the all-new Yahoo! Mail . "The<BR>> New Version is radically easier to use" – The Wall Street Journal<BR></BLOCKQUOTE><BR><p> 
<hr size=1>
Try the <a href="http://us.rd.yahoo.com/mail/uk/taglines/default/nowyoucan/wall_st_2/*http://us.rd.yahoo.com/evt=40565/*http://uk.docs.yahoo.com/nowyoucan.html">all-new Yahoo! Mail</a> . "The New Version is radically easier to use" – The Wall Street Journal