# [FFmpeg-devel] [OPW] OPW Project Proposal

Pallavi Kumari pallavikumari.in at gmail.com
Fri Nov 4 17:31:22 EET 2016

```>> The frequency components after a R->C FFT are complex numbers
>> the peak values of complex numbers are not the same as the maxima of
>> their real components
>> did the paper this is based on say anything about what is recommanded
>> to take here for the maxima?

>> It would seem logic to me to take the magnitude of the complex numbers
>> instead of the real components individually, but that may or may not
>> be what the paper meant
>> not that we have to do what the paper says, we can in fact do something
>> entirely different if it works better

We need points who have high peak intensity(amplitude). Hence, we require
amplitude value at frequency points (say x-axis is freq and y is
amplitude.) so, I guess taking magnitude of complex numbers is logical.
That's how people are using.

>> This prints the amplitude of the peaks, so if the volume is changed
>> to half, all points change to half
>> Wasnt the paper about the locations of the peaks (which would not
>> change if the volume is changed) ?

(say x-axis is freq and y is amplitude.) We will be creating bins for ex:
0-40, 40-80, 80-120, 120-180, 180-300 frequency values. And for each freq
bin interest is to find a points with highest amplitude.

A example set(interest point) is  37, 72, 120, 158

From each window an interest point is extracted.

On Fri, Nov 4, 2016 at 6:41 PM, Michael Niedermayer <michael at niedermayer.cc>
wrote:

> On Fri, Nov 04, 2016 at 05:27:34PM +0530, Pallavi Kumari wrote:
> > patch attached.
> >
> > On Fri, Nov 4, 2016 at 3:29 PM, Michael Niedermayer
> <michael at niedermayer.cc>
> > wrote:
> >
> > > On Fri, Nov 04, 2016 at 07:46:34AM +0530, Pallavi Kumari wrote:
> > > > Updated patch attached. PFA
> > > [...]
> > >
> > > >  Makefile        |    1
> > > >  af_peakpoints.c |  263 ++++++++++++++++++++++++++++++
> > > ++++++++++++++++++++++++++
> > > >  allfilters.c    |    1
> > > >  version.h       |    2
> > > >  4 files changed, 266 insertions(+), 1 deletion(-)
> > > > 7f6affa8dcb632cdea162553a18ab1489f0783ce  0001-avfilter-added-
> > > peakpoints-filter.patch
> > > > From 46be941c87713f8afee686eed0262ca59a2896fd Mon Sep 17 00:00:00
> 2001
> > > > From: Atana <atana at openmailbox.org>
> > > > Date: Fri, 4 Nov 2016 07:43:29 +0530
> > > > Subject: [PATCH] avfilter: added peakpoints filter
> > > >
> > > > ---
> > > >  libavfilter/Makefile        |   1 +
> > > >  libavfilter/af_peakpoints.c | 263 ++++++++++++++++++++++++++++++
> > > ++++++++++++++
> > > >  libavfilter/allfilters.c    |   1 +
> > > >  libavfilter/version.h       |   2 +-
> > > >  4 files changed, 266 insertions(+), 1 deletion(-)
> > > >  create mode 100644 libavfilter/af_peakpoints.c
> > > >
> > > > diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> > > > index 7ed4696..1a18902 100644
> > > > --- a/libavfilter/Makefile
> > > > +++ b/libavfilter/Makefile
> > > > @@ -96,6 +96,7 @@ OBJS-\$(CONFIG_LADSPA_FILTER)                 +=
> > > >  OBJS-\$(CONFIG_LOUDNORM_FILTER)               += af_loudnorm.o
> > > >  OBJS-\$(CONFIG_LOWPASS_FILTER)                += af_biquads.o
> > > >  OBJS-\$(CONFIG_PAN_FILTER)                    += af_pan.o
> > > > +OBJS-\$(CONFIG_PEAKPOINTS_FILTER)             += af_peakpoints.o
> > > >  OBJS-\$(CONFIG_REPLAYGAIN_FILTER)             += af_replaygain.o
> > > >  OBJS-\$(CONFIG_RESAMPLE_FILTER)               += af_resample.o
> > > >  OBJS-\$(CONFIG_RUBBERBAND_FILTER)             += af_rubberband.o
> > > > diff --git a/libavfilter/af_peakpoints.c
> b/libavfilter/af_peakpoints.c
> > > > new file mode 100644
> > > > index 0000000..da108ca
> > > > --- /dev/null
> > > > +++ b/libavfilter/af_peakpoints.c
> > > > @@ -0,0 +1,263 @@
> > > > +/*
> > > > + * Copyright (c) 2016 Atana
> > > > + *
> > > > + * This file is part of FFmpeg.
> > > > + *
> > > > + * FFmpeg is free software; you can redistribute it and/or
> > > > + * modify it under the terms of the GNU Lesser General Public
> > > > + * version 2.1 of the License, or (at your option) any later
> version.
> > > > + *
> > > > + * FFmpeg is distributed in the hope that it will be useful,
> > > > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > > > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> > > > + * Lesser General Public License for more details.
> > > > + *
> > > > + * You should have received a copy of the GNU Lesser General Public
> > > > + * License along with FFmpeg; if not, write to the Free Software
> > > > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> > > 02110-1301 USA
> > > > + */
> > > > +
> > > > +#include "libavcodec/avcodec.h"
> > > > +#include "libavcodec/avfft.h"
> > > > +#include "libavformat/avformat.h"
> > > > +#include "libswscale/swscale.h"
> > > > +#include "avfilter.h"
> > > > +#include "audio.h"
> > > > +#include "libavutil/opt.h"
> > > > +
> > > > +#define SIZECHECK 4096
> > > > +
> > > > +/* Structure to contain peak points context */
> > > > +typedef struct {
> > > > +    const AVClass *class;
> > > > +    double *data;
> > > > +    int nsamples;
> > > > +    int index;
> > > > +    int isOnce;
> > > > +    int buffFlag;
> > > > +    double *peaks;
> > > > +    int size; // number of peaks
> > > > +    int windowSize;
> > > > +    //char *inputFile;
> > > > +} PeakPointsContext;
> > > > +
> > > > +/* returns maximum value from an array conditioned on start and end
> > > index */
> > > > +static double getMax(double *res_arr, int startIndex, int endIndex)
> {
> > > > +    int i;
> > > > +    double max = res_arr[startIndex];
> > > > +    for (i = startIndex; i <= endIndex; i++) {
> > > > +         if (res_arr[i] > max) {
> > > > +             max = res_arr[i];
> > > > +         }
> > > > +    }
> > >
> > > tabs are forbidden in ffmpeg git
> > >
> > >
> > > > +    return max;
> > > > +}
> > > > +
> > > > +/* Stores peak frequency for each window(of chunkSize) in peaks
> array */
> > > > +static void getPeakPointInChunk(int chunkSize, double *res_arr, int
> > > size, double *peaks) {
> > > > +    int i = 0, peakIndex = 0;
> > > > +    int startIndex = 0;
> > > > +    double max;
> > > > +    // get a chunk and find max value in it
> > > > +    while (i < size) {
> > > > +         if (i % chunkSize-1 == 0) {
> > > > +            max = getMax(res_arr, startIndex, i);
> > > > +             peaks[peakIndex++] = max;
> > > > +             startIndex = startIndex + chunkSize;
> > > > +         }
> > > > +        i += 1;
> > > > +    }
> > > > +}
> > > > +
> > > > +/* Get peaks points from windowed frequency domain data*/
> > > > +static int getPeakPoints(PeakPointsContext *ppc) {
> > > > +    int i, m, k, size, chunkSize, pSize, chunkSampleSize, resSize;
> > > > +    double *fft_res;
> > > > +    void *avc;
> > > > +    RDFTContext *rdftC;
> > > > +    FFTSample *data;
> > > > +
> > > > +    size = ppc->index;
> > > > +    m = log2(ppc->windowSize);
> > > > +    chunkSize = ppc->windowSize;
> > > > +    chunkSampleSize = size/chunkSize;
> > > > +    resSize = chunkSize * chunkSampleSize;
> > > > +
> > >
> > > > +    fft_res = av_malloc_array(resSize, sizeof(double));
> > >
> > > This seems never freed
> > >
> > >
> > > > +
> > > > +    if (!fft_res) {
> > > > +        av_log(avc, AV_LOG_ERROR, "Cann't allocate memmory for
> storing
> > > fft data\n");
> > > > +        return 0;
> > > > +    }
> > > > +
> > > > +
> > > > +    rdftC = av_rdft_init(m, DFT_R2C);
> > > > +    data = av_malloc_array(chunkSize, sizeof(FFTSample));
> > > > +
> > > > +    if (!data) {
> > > > +        av_log(avc, AV_LOG_ERROR, "Cann't allocate memmory for chunk
> > > fft data\n");
> > > > +        return 0;
> > > > +    }
> > > > +    // FFT transform for windowed time domain data
> > > > +    // window is of size chunkSize
> > > > +    k = 0;
> > > > +    while (k < resSize) {
> > > > +        //copy data
> > > > +        for (i = 0; i < chunkSize; i++) {
> > > > +            data[i] = ppc->data[i+k];
> > > > +        }
> > > > +        //calculate FFT
> > > > +        av_rdft_calc(rdftC, data);
> > > > +        for (i = 0; i < chunkSize; i++) {
> > > > +         fft_res[i+k] = data[i];
> > > > +        }
> > > > +        k = k + chunkSize;
> > > > +    }
> > > > +
> > > > +    av_rdft_end(rdftC);
> > > > +    pSize = resSize/chunkSize;
> > > > +    ppc->size = pSize;
> > >
> > > > +    ppc->peaks = av_malloc_array(pSize, sizeof(double));
> > >
> > > This is allocated potentially durng each filter_frame
> > > but only deallocated once at the end, all except the last thus leak
> > >
> > >
> > > [...]
> > > > +{
> > > > +    AVFilterContext *ctx = inlink->dst;
> > > > +    PeakPointsContext *p = ctx->priv;
> > > > +    int i, nb_samples = samples->nb_samples;
> > > > +
> > > > +    // store audio data
> > > > +    for (i = 0; i < nb_samples; i++) {
> > >
> > > > +        p->data[p->index] = (double)samples->data[0][i];
> > >
> > > This reads the double values as 8bit integers, you have to cast to
> > > a pointer to double and dereference it to read the double values
> > >
> > >
> > > [...]
> > >
> > >
> > > --
> > > Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC7
> 87040B0FAB
> > >
> > > Breaking DRM is a little like attempting to break through a door even
> > > though the window is wide open and the only thing in the house is a
> bunch
> > > of things you dont want and which you would get tomorrow for free
> anyway
> > >
> > > _______________________________________________
> > > ffmpeg-devel mailing list
> > > ffmpeg-devel at ffmpeg.org
> > > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> > >
> > >
>
> >  Makefile        |    1
> >  af_peakpoints.c |  267 ++++++++++++++++++++++++++++++
> ++++++++++++++++++++++++++
> >  allfilters.c    |    1
> >  version.h       |    2
> >  4 files changed, 270 insertions(+), 1 deletion(-)
> peakpoints-filter.patch
> > From 5b7efa32b1e3970d6accbb23fee02331278b118b Mon Sep 17 00:00:00 2001
> > From: Atana <atana at openmailbox.org>
> > Date: Fri, 4 Nov 2016 07:43:29 +0530
> > Subject: [PATCH] avfilter: added peakpoints filter
> >
> > ---
> >  libavfilter/Makefile        |   1 +
> >  libavfilter/af_peakpoints.c | 267 ++++++++++++++++++++++++++++++
> ++++++++++++++
> >  libavfilter/allfilters.c    |   1 +
> >  libavfilter/version.h       |   2 +-
> >  4 files changed, 270 insertions(+), 1 deletion(-)
> >  create mode 100644 libavfilter/af_peakpoints.c
> >
> > diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> > index 7ed4696..1a18902 100644
> > --- a/libavfilter/Makefile
> > +++ b/libavfilter/Makefile
> > @@ -96,6 +96,7 @@ OBJS-\$(CONFIG_LADSPA_FILTER)                 +=
> >  OBJS-\$(CONFIG_LOUDNORM_FILTER)               += af_loudnorm.o
> >  OBJS-\$(CONFIG_PAN_FILTER)                    += af_pan.o
> > +OBJS-\$(CONFIG_PEAKPOINTS_FILTER)             += af_peakpoints.o
> >  OBJS-\$(CONFIG_REPLAYGAIN_FILTER)             += af_replaygain.o
> >  OBJS-\$(CONFIG_RESAMPLE_FILTER)               += af_resample.o
> >  OBJS-\$(CONFIG_RUBBERBAND_FILTER)             += af_rubberband.o
> > diff --git a/libavfilter/af_peakpoints.c b/libavfilter/af_peakpoints.c
> > new file mode 100644
> > index 0000000..cb64b82
> > --- /dev/null
> > +++ b/libavfilter/af_peakpoints.c
> > @@ -0,0 +1,267 @@
> > +/*
> > + * Copyright (c) 2016 Atana
> > + *
> > + * This file is part of FFmpeg.
> > + *
> > + * FFmpeg is free software; you can redistribute it and/or
> > + * modify it under the terms of the GNU Lesser General Public
> > + * version 2.1 of the License, or (at your option) any later version.
> > + *
> > + * FFmpeg is distributed in the hope that it will be useful,
> > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> > + * Lesser General Public License for more details.
> > + *
> > + * You should have received a copy of the GNU Lesser General Public
> > + * License along with FFmpeg; if not, write to the Free Software
> > + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
> 02110-1301 USA
> > + */
> > +
> > +#include "libavcodec/avcodec.h"
> > +#include "libavcodec/avfft.h"
> > +#include "libavformat/avformat.h"
> > +#include "libswscale/swscale.h"
> > +#include "avfilter.h"
> > +#include "audio.h"
> > +#include "libavutil/opt.h"
> > +
> > +#define SIZECHECK 4096
> > +
> > +/* Structure to contain peak points context */
> > +typedef struct {
> > +    const AVClass *class;
> > +    double *data;
> > +    int nsamples;
> > +    int index;
> > +    int isOnce;
> > +    int buffFlag;
> > +    double *peaks;
> > +    int size; // number of peaks
> > +    int windowSize;
> > +    //char *inputFile;
> > +} PeakPointsContext;
> > +
> > +/* returns maximum value from an array conditioned on start and end
> index */
> > +static double getMax(double *res_arr, int startIndex, int endIndex) {
> > +    int i;
> > +    double max = res_arr[startIndex];
> > +    for (i = startIndex; i <= endIndex; i++) {
> > +         if (res_arr[i] > max) {
> > +             max = res_arr[i];
> > +         }
> > +    }
> > +    return max;
> > +}
> > +
> > +/* Stores peak frequency for each window(of chunkSize) in peaks array */
> > +static void getPeakPointInChunk(int chunkSize, double *res_arr, int
> size, double *peaks) {
> > +    int i = 0, peakIndex = 0;
> > +    int startIndex = 0;
> > +    double max;
> > +    // get a chunk and find max value in it
> > +    while (i < size) {
> > +         if (i % chunkSize-1 == 0) {
> > +            max = getMax(res_arr, startIndex, i);
> > +            peaks[peakIndex++] = max;
> > +            startIndex = startIndex + chunkSize;
> > +       }
> > +       i += 1;
> > +    }
> > +}
>
> The frequency components after a R->C FFT are complex numbers
> the peak values of complex numbers are not the same as the maxima of
> their real components
> did the paper this is based on say anything about what is recommanded
> to take here for the maxima?
>
> It would seem logic to me to take the magnitude of the complex numbers
> instead of the real components individually, but that may or may not
> be what the paper meant
> not that we have to do what the paper says, we can in fact do something
> entirely different if it works better
>
> [...]
>
> > +static void ppointsStats(AVFilterContext *ctx, PeakPointsContext *p) {
> > +    int i, ret;
> > +    ret = getPeakPoints(p);
> > +
> > +    if (ret && p->size) {
> > +     // print peaks
> > +        if (p->isOnce) {
> > +            av_log(ctx, AV_LOG_INFO, "######## Peak points are
> ########\n");
> > +            p->isOnce = 0;
> > +        }
> > +        for (i = 0; i < p->size; i++) {
> > +            av_log(ctx, AV_LOG_INFO, "%f\n", p->peaks[i]);
> > +        }
> > +        // free peaks and set size to zero
> > +        av_freep(&p->peaks);
> > +        p->size = 0;
> > +    } else if (p->size || !ret) {
> > +        av_log(ctx, AV_LOG_ERROR, "Peak points not retrieved\n");
> > +        return;
> > +    }
> > +}
>
> This prints the amplitude of the peaks, so if the volume is changed
> to half, all points change to half
> Wasnt the paper about the locations of the peaks (which would not
> change if the volume is changed) ?
>
> [...]
> --
>
> If you think the mosad wants you dead since a long time then you are either
> wrong or dead since a long time.
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>
>
```