That file looks like an earlier change, did you send the right file ?
--
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast.
Thank you Stephen for all that hard work! I have looked at the patch
and made a few changes:
+ I'd like the keep the lpc order argument in lpc.c
+ I'd like to keep the VQ dimension flexible in quantise.c
I've attached the patch with my edits (ie I removed a few hunks as per
above). Could you pl
Ack, OK I'm awake now, I forgot you requested the -ruN diff.
I did a:
diff -ruN freetel-code/codec2-dev/src freetel-code-new/codec2-dev/src
>patch.txt
and put the patch at:
http://goo.gl/4UZAFX
--
HPCC Systems Open Sour
Here's a diff file of the changes I made to codec2-dev from revision 1700
(probably too big a file for email).
http://goo.gl/J2qs9C
I tested it against fdmdv2 (I'm not able to get fdmdv2-dev working yet) and
it works with it. I tried a few c2sim and c2enc/c2dec options, and they
worked.
I did ge
oof, sorry, I thought the codec2-dev was your playground. I get it now.
Yes, let me start over with that and I'll check the c2sim stuff as well.
--
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Hello Steve,
Thank you for all that work. In the past I was messing with the LPC
order, but I think it's safe to say it's fixed at 10 now. It will also
simply the code which is nice.
Can you please:
1/ Confirm c2sim will still allow switching the lpc modelling on and off
with the --lpc command
I tested the previous change by linking it to fdmdv2 build and freedv does
still work.
One thing I've noticed is that the SNR is a bit lower (27 - 28). I didn't
get any bit failures during a test audio receive.
I did note that we have a new FIR filter for the pilot carrier, and it
seems to be mec
Actually there is a STM32F4 library fft that I tried last year when
porting Codec 2 to the STM32F4. It was actually a little slower than
kiss fft!
Based on my STM32F4 experience (i.e. Codec 2 runs just fine), the fft
memory isn't a big problem ATM. What _is_ a problem is the memory
consumed by s
It would be nice to find an ARM Cortex-M4 floating point *inline* FFT.
That kiss-fft eats memory like I eat BBQ pork products...
Alas, I've been out of embedded processors for way too long.
--
HPCC Systems Open Source Big
Thanks - that's exactly what I need :-)
On Thu, 2014-06-19 at 18:32 -0500, Steve Sampson wrote:
> try this. Not sure about the -ruN option. I was just editing the
> original.
> --
> HPCC Systems Open Source Big Data Platf
try this. Not sure about the -ruN option. I was just editing the original.
Index: codec2/src/quantise.c
===
--- codec2/src/quantise.c (revision 1692)
+++ codec2/src/quantise.c (working copy)
@@ -795,7 +795,7 @@
COMP
David,
Did you get the "bits in int arrays" issue? It's easy enough to make GET
and SET macros that do the proper bit packing.
Thanks
Bruce
--
HPCC Systems Open Source Big Data Platform from LexisNexis Risk S
Thanks Steve I've checked it into codec2-dev, nice catch.
If you could use diff -RuN and attach the patch as a txt file next time
that would be much appreciated - my email client mangles the spaces.
Thanks,
David
On Thu, 2014-06-19 at 17:07 -0500, Steve Sampson wrote:
> Here's some data:
>
>
Here's some data:
$ svn diff
Index: codec2/src/quantise.c
===
--- codec2/src/quantise.c (revision 1692)
+++ codec2/src/quantise.c (working copy)
@@ -795,7 +795,7 @@
COMP Ww[FFT_ENC]; /* weighting spectrum */
floa
Thanks Steve,
Looks good:
1/ Can you send me the change in patch format please?
2/ Have you tested the change to ensure the output decoded speech is
identical (e.g. using diff)?
Thanks,
David
On Thu, 2014-06-19 at 09:52 -0500, Steve Sampson wrote:
> As a translater I usually have to at leas
As a translater I usually have to at least look at the algorithms, so I
thought I'd find another 512 floats and replace it with 1 float.
in module quantise.c
In function lpc_post_filter()
The Pfw array is not used after being created. Making it a variable saves
2k bytes.
Change:
float Pfw[FFT_
16 matches
Mail list logo
