Greg - LPCNet quantised to 3000 kbit/s would indeed be an interesting
comparison; Jean-Marc and I have been quantising LPCNet at beneath 2000
bit/s, in my case to squeeze it through a HF Radio channel. I suspect
the focus on 3000 bit/s is because that is the lowest rate that makes
sense for VOIP.
Adrian - there is a paper linked from the Lyra page.
- David
On 28/2/21 11:09 pm, Adrian Musceac wrote:
While interesting and newsworthy, I'd assume from the start that this
codec has the same advantages and pitfalls as other ML applications,
i.e. works very well in 90% of cases and fails dramatically in 10% of
edge cases. Even though the page specifies it is aimed at a completely
different domain (not radio) I would say there is no chance it can
replace Codec2 soon, and I think the same for LPCNet unfortunately.
Sometimes 99.9 % of reliability of a robotic voice is better than 90%
reliability of high quality voice. Amateur radio seems to me like a
combination of all kinds of languages, foreign accents and all sorts
of messy real world input.
That said, I'd still be interested to read a paper regarding their
approach and improvements to state of the art.
Adrian
On February 28, 2021 2:16:16 AM UTC, David Rowe <da...@rowetel.com>
wrote:
Hi Michael,
Thanks for the post - very interesting. Sure is an exciting time
for speech coding.
The speech quality of Lyra is vastly better than anything Codec 2
can offer at a similar bit rate (3000 bits/s). Codec 2 would sound
closer to the Speex 3 kbit/s samples on the Lyra page. Lyra has
reasonable CPU complexity (single thread of a modern smartphone),
but that's still much higher than Codec 2 (which runs on
microcontrollers). Lyra presently has no low bit rates modes
(Codec 2 is commonly used at 700 bits/ for HF radio
applications). Codec 2 is open source and GPL Licensed. I'm not
clear where Lyra stands on those issues. The latency of Lyra
appears quite high (90ms).
Lyra is a Machine Learning (ML) based codec, so in a similar class
to Jean-Marc Valin's LPCNet, which is open source. Our initial
attempt with using LPCNet in real world scenarios (e.g. FreeDV
2020 for HF radio) shows a lot of promise, but has some speaker
dependence and possibly quantisation issues (it breaks down on
some speakers - sounds great on others). The Lyra team claim to
have put a lot of work into speaker-independence, and the project
has a lot of resources behind it :-)
Codec 2 presently has one very part-time person working on the
core speech codec and several other people kindly contributing to
other parts of Codec 2/FreeDV :-)
At 3kbit/s Lyra would fit neatly into VHF/UHF radio type
applications, that would be a cool demo.
Cheers,
David
On 28/2/21 4:50 am, mgraves mstvp.com wrote:
Hi,
While I’ve watched from afar, this is my first message to this list.
I was wondering how Codec 2 compares to this latest effort from
Google; Lyra.
https://ai.googleblog.com/2021/02/lyra-new-very-low-bitrate-codec-for.html?m=1
Michael Graves
mgra...@mstvp.com <mailto:mgra...@mstvp.com>
o: (713) 861-4005
c: (713) 201-1262
sip:mgra...@mjg.onsip.com
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
