Dear David??
Any new progress about Codec2, the pitch estimator algorithm or the
voiced/unvoiced classification algorithm ?
Best regards.
Hua
------------------ Original ------------------
From:
"freetel-codec2"
<da...@rowetel.com>;
Date: Tue, Sep 14, 2021 04:49 PM
To: "freetel-codec2"<freetel-codec2@lists.sourceforge.net>;
Subject: Re: [Freetel-codec2] facebook speech codec at 365bps
On Tue, 2021-09-14 at 09:24 +0800, Random via Freetel-codec2 wrote:
> In my opinion, the pitch estimator is the bottleneck of any
> narrowband vocoder, no matter the tranditional or the neural-network
> based.
>
> LPCnet uses the (last-century) RAPT algorithm for the pitch
> estimation, and if the pitch value is wrong, the output would be very
> terrible.
>
A few years ago I spent some time exploring LPCNet, including working
with a few different pitch estimators. I found that, quite remarkably,
LPCnet was insensitive to gross errors in the pitch estimator. This
really surprised me, as I have a history of pain in developing pitch
estimators. They are critical to old school low bit rate codecs like
Codec 2, MELP, AMBE.
Lyra doesn't even have a pitch estimator - just sends high order mel-
spaced speech spectra, which I guess is densely sampled enough to
resolve pitch harmonics.
So my experience is NN codecs can bypass the harder parts of the whole
pitch estimator issue ... I think they are also picking up other cues
like the presence of low frequency harmonics in the mel spectra for
males.
In fact some of our early quantisation attempts at LPCNet had errors
like males turning into females (pitch shifts) if we got the spectral
quantisation at low frequency wrong ;-)
An interesting research question - is a pitch "feature" needed for a
low bit rate NN codec? LPCNet has one (1600 bits/s) - Lyra doesn't
(3000 bits/s).
- David
>
> ------------------ Original ------------------
> From: "freetel-codec2" <da...@rowetel.com>;
> Date: Tue, Sep 14, 2021 06:25 AM
> To: "freetel-codec2"<freetel-codec2@lists.sourceforge.net>;
> Subject: Re: [Freetel-codec2] facebook speech codec at 365bps
>
> On Mon, 2021-09-13 at 07:24 +0000, Greg Maxwell wrote:
> > On Mon, Sep 13, 2021 at 7:05 AM Random via Freetel-codec2
> > <freetel-codec2@lists.sourceforge.net> wrote:
> > > Is it speaker-independent ?
> >
> > It's speaker independent with the additional per-speaker data
> > mentioned in my post.
> >
>
> That sounds like speaker dependence to me.
>
> I encountered this with the early LPCNet work as well (as used in
> FreeDV 2020), the quality dropped off significantly for about 10% of
> voices (including mine!). However I haven't tried the latest version
> of LPCnet from Jean-Marc, he's been steadily improving his NN model
> and
> codec.
>
> The Lyra paper mentions some specific work in this area, so I'm sure
> it
> will be addressed in time. High quality, speaker independent speech
> coding at sub 1000 bit's certainly feels possible.
>
> Another issue to address is robustness to bit errors. In codec 2 I
> avoid inter-frame coding (ie coding differences) to keep some
> tolerance
> to the high bit error rates. This costs a few bits/s compared to a
> super efficient approach.
>
> I figure tolerance to bit errors might be something we can train for
> in
> NN codecs.
>
> - David
>
>
>
>
> _______________________________________________
> 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_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
