Hi Jeroen and list,
I have broken up Jeroen's sample below into separate sample files, and
for comparison added the same samples processed by Codec 2 700C and 1300:
http://rowetel.com/downloads/codec2/jeroen_rfc_1300c.tar.gz
This way each sample can be compared to 700C and 1300. BTW the script
to separate all.wav into separate files is:
codec2-dev/scripts/separate_all.sh.
When I compare codecs, I take one (say 3 second) sample and listen to it
processed by different codecs. I have attached a screen shot of how I
organise my file manager for these samples.
What do you all think of Jeroen's 1300 candidate compared to 700C and
1300. Is it superior to the current 1300, or are they roughly the same?
Which samples work well, which ones are problematic?
In this way we look at Jeroens fine work, find out which techniques
work, and where some more work is required.
Crowd-sourced codec evaluation......
Thanks,
David
On 04/05/17 09:40, Jeroen Vreeken wrote:
Hi all,
It has been a while, but I did make some steady progress.
(Trying different training strategies can take days at a time)
I made a new version of the patch with some changes:
- I cut out a lot of silence and noises from the training material (like
the breathing). There is no real point in reproducing such noises
perfectly. I also added David's samples and some other nice ones I found
that where different from all I already had. (Also some children voices)
- Instead of three times 256 values (resulting in 24 bits) I used two
tables of 1024 values. (20bits). 24 bits of course give a slightly
better result, but I think the bits are better used now. On average the
error is about 1.7dB lower than 700C.
- Wo and Energy each get an extra bit.
- The last two bits are used for voicing. One of the things I noticed
when listening to audible artifacts was that the combined Wo and voicing
was resulting in Wo being interpolated from the hardcoded non-voiced
default back to the voiced value. This sometimes results in an audible
pitch shift. Having a voice bits seperatly makes this a bit less likely.
Sample:
http://video.vreeken.net/~dml/patches/all.1300C_10b10b5e7wo2v.wav
73,
Jeroen PE1RXQ
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
