[FFmpeg-devel] [PATCH] libswresample: avoid s16p internal processing format

[Paul B Mahol]

On 1/8/23, Michael Niedermayer <michael at niedermayer.cc> wrote:
> On Fri, Jan 06, 2023 at 07:04:59PM +0100, Paul B Mahol wrote:
>> On Fri, Jan 6, 2023 at 7:01 PM Paul B Mahol <onemda at gmail.com> wrote:
>>
>> >
>> >
>> > On Fri, Jan 6, 2023 at 6:25 PM Michael Niedermayer
>> > <michael at niedermayer.cc>
>> > wrote:
>> >
>> >> On Thu, Jan 05, 2023 at 11:08:25PM +0100, Paul B Mahol wrote:
>> >> > On Thu, Jan 5, 2023 at 9:53 PM Michael Niedermayer <
>> >> michael at niedermayer.cc>
>> >> > wrote:
>> >> >
>> >> > > On Thu, Jan 05, 2023 at 01:44:10PM +0100, Paul B Mahol wrote:
>> >> > > > Patch attached.
>> >> > >
>> >> > > >  swresample.c |    3 ++-
>> >> > > >  1 file changed, 2 insertions(+), 1 deletion(-)
>> >> > > > eee7a0685b44aa867562138a2e2437ecb8844612
>> >> > > 0001-libswresample-swresample-avoid-s16p-internal-transfe.patch
>> >> > > > From 9c4cd60e2dd41cf98d693c8251f4cfade0807073 Mon Sep 17
>> >> > > > 00:00:00
>> >> 2001
>> >> > > > From: Paul B Mahol <onemda at gmail.com>
>> >> > > > Date: Thu, 5 Jan 2023 13:40:12 +0100
>> >> > > > Subject: [PATCH] libswresample/swresample: avoid s16p internal
>> >> transfer
>> >> > > format
>> >> > > >
>> >> > > > Instead use float one by default for sample rate conversions.
>> >> > > > The s16p internal transfer format produces visible and hearable
>> >> > > > quantization artifacts.
>> >> > >
>> >> > > When does this occur and why?
>> >> > >
>> >> >
>> >> > It occurs always. Just compare output with 16bit and
>> >> > int32/float/double.
>> >> > Look at other people report on internet.
>> >> > Look at src.infinitewave.ca
>> >>
>> >> src.infinitewave.ca uses 32bit none of what it shows should touch the
>> >> codepath
>> >> you change.
>> >>
>> >> if we look at src.infinitewave.ca for swr we see 2 types of artifacts
>> >> 1. Aliassing which is at maybe -120db with the actual signal at 0db
>> >>    i would like to see some evidence that a human can hear this
>> >>
>> >
>> > For s16p<->s16p it is much lower, around -78dB thus this patch.
>> >
>> > Also for others and reports for swr its is lower than exact -120dB
>> >
>> >
>> > 2. Reflection and attenuation at the transition frequency
>> >> With linear filters there is a tradeof between attenuation of the
>> >> passband, reflection of frequencies beyond, latency and so on
>> >> You can have a perfect sharp cutoff with no attenuation and no
>> >> refelection
>> >> that requires a infinitly long filter. And while this looks best in
>> >> this
>> >> frequency plot, does it actually sound best ? If you can hear -120db
>> >> signals you surely would then also hear the ringing long before a
>> >> gunshot
>> >> from such long filter.
>> >>
>> >
>> One can always change linear FIR to be min phase FIR kernel.
>
> I certainly would welcome a wider range of filters in swr, if you want to
> add
> any low delay sinc approximation or in fact i would welcome any filter
> you want to add.

I have hack for simple cases when one of decimation/interpolation factor is 1
than there is single only FIR kernel applied and then one do not use
naive convolution which is O(N^2) but O(Nlog(N)) when using RDFT/FFT.

For bigger filter sizes it makes filtering amazingly faster. (from
<80x to >400x).

So I'm interested in making it faster if possible for any combination
of decimation/interpolation.

>
> thx
>
> [...]
> --
> Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
>
> The real ebay dictionary, page 1
> "Used only once"    - "Some unspecified defect prevented a second use"
> "In good condition" - "Can be repaird by experienced expert"
> "As is" - "You wouldnt want it even if you were payed for it, if you knew
> ..."
>