[FFmpeg-devel] [PATCH] avfilter: add arbitrary audio FIR filter

[Paul B Mahol]

On 5/6/17, Muhammad Faiz <mfcc64 at gmail.com> wrote:
> On Sat, May 6, 2017 at 2:33 AM, Paul B Mahol <onemda at gmail.com> wrote:
>> On 5/5/17, Muhammad Faiz <mfcc64 at gmail.com> wrote:
>>>>>> Is pts handled correctly here? Seem it is not derived from input pts.
>>>>>>
>>>>>
>>>>> It can not be derived in any other way.
>>>>
>>>> Probably, at least, first pts should be derived from input pts.
>>>> Also, is time_base always 1/sample_rate?
>>>>
>>>> Thank's.
>>>
>>> Probably, like in asetnsamples filter.
>>
>> Done. Have an idea where artifacst come out for some IRs?
>
> I have no idea what's wrong here? I currently don't understand
> partitioned convolution.
>
> My test case:
> ./ffplay -f lavfi "aevalsrc='if(lt(t,1), if(n,0,1), sin(1000*t*t))',
> aformat= channel_layouts=mono, asplit [afir_in0], firequalizer=
> zero_phase=on:gain_entry='entry(0, 0); entry(100, -4); entry(1000,
> -30); entry(5000, -50); entry(6000, -10); entry(10000, -5)', aformat=
> channel_layouts=mono,a
> split [merge1], atrim= end_sample=883 [afir_in1]; [afir_in0][afir_in1]
> afir, aformat=channel_layouts=mono, [merge1] amerge,
> asplit[out0],showspectrum=s=1024x512[out1]"
>
> Note that your old patch didn't generate artifact.

Artifacts are gone if you use bigger IR size, >2048.

UPOLS should be rather trivial to understand and to implement.