[FFmpeg-devel] [PATCH] avfilter: add arnndn filter

[Paul B Mahol]

On Sat, Jul 27, 2019 at 3:55 PM Lynne <dev at lynne.ee> wrote:

> Jul 26, 2019, 4:37 PM by onemda at gmail.com <mailto:onemda at gmail.com>:
> > +static void biquad(float *y, float mem[2], const float *x,
> > +                   const float *b, const float *a, int N)
> > +{
> > +    for (int i = 0; i < N; i++) {
> > +        float xi, yi;
> > +        xi = x[i];
> > +        yi = x[i] + mem[0];
> > +        mem[0] = mem[1] + (b[0]*(double)xi - a[0]*(double)yi);
> > +        mem[1] = (b[1]*(double)xi - a[1]*(double)yi);
> > +        y[i] = yi;
> > +    }
> > +}
>
> Do you really need double precision here?
>
>
> > +static void apply_window(float *x, float *half_window)
> > +{
> > +    for (int i = 0; i < FRAME_SIZE; i++) {
> > +        x[i] *= half_window[i];
> > +        x[WINDOW_SIZE - 1 - i] *= half_window[i];
> > +    }
> > +}
>
> Use vector_fmul_window.
>
>
> > +#define FRAME_SIZE_SHIFT 2
> > +#define FRAME_SIZE (120<<FRAME_SIZE_SHIFT)
> > +#define WINDOW_SIZE (2*FRAME_SIZE)
> > +#define FREQ_SIZE (FRAME_SIZE + 1)
> >
> > +    AVComplexFloat y[WINDOW_SIZE];
> >
> > +    st->tx_fn(st->tx, y, x, 0);
> > +
> > +    for (int i = 0; i < FREQ_SIZE; i++) {
> > +        out[i] = y[i];
> > +    }
>
> Make y and x aligned and set the tx stride to sizeof(float) for when SIMD
> is implemented.
> I can't see how this could even work with stride = 0.
> Why is FREQ_SIZE = FRAME_SIZE + 1?
>

Because how FFT works, the first im and last im is always 0 so very last re
is stored in first as im.
So that output size stays same.


>
>
> > +    st->tx_fn(st->tx, y, x, 0);
> > +
> > +    /* output in reverse order for IFFT. */
> > +    out[0] = y[0].re / WINDOW_SIZE;
> > +    for (i=1; i < WINDOW_SIZE; i++) {
> > +        out[i] = y[WINDOW_SIZE - i].re / WINDOW_SIZE;
> > +    }
>
> Make another tx context for an inverse transform, also same as above, and
> use float_dsp to divide all by WINDOW_SIZE.
>
>
> > +    for (int i=0;i<FRAME_SIZE;i++)
> > +        out[i] = x[i] + st->synthesis_mem[i]
>
> float_dsp? If you can afford to overwrite and overread the arrays odd
> sizes will work.
>
>
> > +    for (int i = 0; i < FRAME_SIZE; i++)
> > +        s->half_window[i] = sin(.5*M_PI*sin(.5*M_PI*(i+.5)/FRAME_SIZE)
> * sin(.5*M_PI*(i+.5)/FRAME_SIZE));
>
> Just copy ff_celt_window and use it.
>
>
> Also fix coding style and upload a weights file somewhere.
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".