Merge ../mlt

[melted] / src / modules / normalize / filter_volume.c

/*
 * filter_volume.c -- adjust audio volume
 * Copyright (C) 2003-2004 Ushodaya Enterprises Limited
 * Author: Dan Dennedy <dan@dennedy.org>
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program 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 General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software Foundation,
 * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
 */

#include <framework/mlt_filter.h>
#include <framework/mlt_frame.h>

#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include <ctype.h>
#include <string.h>

#define MAX_CHANNELS 6
#define EPSILON 0.00001

/* The following normalise functions come from the normalize utility:
   Copyright (C) 1999--2002 Chris Vaill */

#define samp_width 16

#ifndef ROUND
# define ROUND(x) floor((x) + 0.5)
#endif

#define DBFSTOAMP(x) pow(10,(x)/20.0)

/** Return nonzero if the two strings are equal, ignoring case, up to
    the first n characters.
*/
int strncaseeq(const char *s1, const char *s2, size_t n)
{
        for ( ; n > 0; n--)
        {
                if (tolower(*s1++) != tolower(*s2++))
                        return 0;
        }
        return 1;
}

/** Limiter function.
 
         / tanh((x + lev) / (1-lev)) * (1-lev) - lev        (for x < -lev)
         |
    x' = | x                                                (for |x| <= lev)
         |
         \ tanh((x - lev) / (1-lev)) * (1-lev) + lev        (for x > lev)
 
  With limiter level = 0, this is equivalent to a tanh() function;
  with limiter level = 1, this is equivalent to clipping.
*/
static inline double limiter( double x, double lmtr_lvl )
{
        double xp = x;

        if (x < -lmtr_lvl)
                xp = tanh((x + lmtr_lvl) / (1-lmtr_lvl)) * (1-lmtr_lvl) - lmtr_lvl;
        else if (x > lmtr_lvl)
                xp = tanh((x - lmtr_lvl) / (1-lmtr_lvl)) * (1-lmtr_lvl) + lmtr_lvl;

//      if ( x != xp )
//              fprintf( stderr, "filter_volume: sample %f limited %f\n", x, xp );

        return xp;
}


/** Takes a full smoothing window, and returns the value of the center
    element, smoothed.

    Currently, just does a mean filter, but we could do a median or
    gaussian filter here instead.
*/
static inline double get_smoothed_data( double *buf, int count )
{
        int i, j;
        double smoothed = 0;

        for ( i = 0, j = 0; i < count; i++ )
        {
                if ( buf[ i ] != -1.0 )
                {
                        smoothed += buf[ i ];
                        j++;
                }
        }
        smoothed /= j;
//      fprintf( stderr, "smoothed over %d values, result %f\n", j, smoothed );

        return smoothed;
}

/** Get the max power level (using RMS) and peak level of the audio segment.
 */
double signal_max_power( int16_t *buffer, int channels, int samples, int16_t *peak )
{
        // Determine numeric limits
        int bytes_per_samp = (samp_width - 1) / 8 + 1;
        int16_t max = (1 << (bytes_per_samp * 8 - 1)) - 1;
        int16_t min = -max - 1;
        
        double *sums = (double *) calloc( channels, sizeof(double) );
        int c, i;
        int16_t sample;
        double pow, maxpow = 0;

        /* initialize peaks to effectively -inf and +inf */
        int16_t max_sample = min;
        int16_t min_sample = max;
  
        for ( i = 0; i < samples; i++ )
        {
                for ( c = 0; c < channels; c++ )
                {
                        sample = *buffer++;
                        sums[ c ] += (double) sample * (double) sample;
                        
                        /* track peak */
                        if ( sample > max_sample )
                                max_sample = sample;
                        else if ( sample < min_sample )
                                min_sample = sample;
                }
        }
        for ( c = 0; c < channels; c++ )
        {
                pow = sums[ c ] / (double) samples;
                if ( pow > maxpow )
                        maxpow = pow;
        }
                        
        free( sums );
        
        /* scale the pow value to be in the range 0.0 -- 1.0 */
        maxpow /= ( (double) min * (double) min);

        if ( -min_sample > max_sample )
                *peak = min_sample / (double) min;
        else
                *peak = max_sample / (double) max;

        return sqrt( maxpow );
}

/* ------ End normalize functions --------------------------------------- */

/** Get the audio.
*/

static int filter_get_audio( mlt_frame frame, int16_t **buffer, mlt_audio_format *format, int *frequency, int *channels, int *samples )
{
        // Get the properties of the a frame
        mlt_properties properties = MLT_FRAME_PROPERTIES( frame );
        double gain = mlt_properties_get_double( properties, "volume.gain" );
        double max_gain = mlt_properties_get_double( properties, "volume.max_gain" );
        double limiter_level = 0.5; /* -6 dBFS */
        int normalise =  mlt_properties_get_int( properties, "volume.normalise" );
        double amplitude =  mlt_properties_get_double( properties, "volume.amplitude" );
        int i, j;
        double sample;
        int16_t peak;

        // Get the filter from the frame
        mlt_filter this = mlt_properties_get_data( properties, "filter_volume", NULL );

        // Get the properties from the filter
        mlt_properties filter_props = MLT_FILTER_PROPERTIES( this );

        if ( mlt_properties_get( properties, "volume.limiter" ) != NULL )
                limiter_level = mlt_properties_get_double( properties, "volume.limiter" );
        
        // Get the producer's audio
        mlt_frame_get_audio( frame, buffer, format, frequency, channels, samples );
//      fprintf( stderr, "filter_volume: frequency %d\n", *frequency );

        // Determine numeric limits
        int bytes_per_samp = (samp_width - 1) / 8 + 1;
        int samplemax = (1 << (bytes_per_samp * 8 - 1)) - 1;
        int samplemin = -samplemax - 1;

        if ( normalise )
        {
                int window = mlt_properties_get_int( filter_props, "window" );
                double *smooth_buffer = mlt_properties_get_data( filter_props, "smooth_buffer", NULL );

                if ( window > 0 && smooth_buffer != NULL )
                {
                        int smooth_index = mlt_properties_get_int( filter_props, "_smooth_index" );
                        
                        // Compute the signal power and put into smoothing buffer
                        smooth_buffer[ smooth_index ] = signal_max_power( *buffer, *channels, *samples, &peak );
//                      fprintf( stderr, "filter_volume: raw power %f ", smooth_buffer[ smooth_index ] );
                        if ( smooth_buffer[ smooth_index ] > EPSILON )
                        {
                                mlt_properties_set_int( filter_props, "_smooth_index", ( smooth_index + 1 ) % window );

                                // Smooth the data and compute the gain
//                              fprintf( stderr, "smoothed %f over %d frames\n", get_smoothed_data( smooth_buffer, window ), window );
                                gain *= amplitude / get_smoothed_data( smooth_buffer, window );
                        }
                }
                else
                {
                        gain *= amplitude / signal_max_power( *buffer, *channels, *samples, &peak );
                }
        }
        
//      if ( gain > 1.0 && normalise )
//              fprintf(stderr, "filter_volume: limiter level %f gain %f\n", limiter_level, gain );

        if ( max_gain > 0 && gain > max_gain )
                gain = max_gain;

        // Initialise filter's previous gain value to prevent an inadvertant jump from 0
        mlt_position last_position = mlt_properties_get_position( filter_props, "_last_position" );
        mlt_position current_position = mlt_frame_get_position( frame );
        if ( mlt_properties_get( filter_props, "_previous_gain" ) == NULL
             || current_position != last_position + 1 )
                mlt_properties_set_double( filter_props, "_previous_gain", gain );

        // Start the gain out at the previous
        double previous_gain = mlt_properties_get_double( filter_props, "_previous_gain" );

        // Determine ramp increment
        double gain_step = ( gain - previous_gain ) / *samples;
//      fprintf( stderr, "filter_volume: previous gain %f current gain %f step %f\n", previous_gain, gain, gain_step );

        // Save the current gain for the next iteration
        mlt_properties_set_double( filter_props, "_previous_gain", gain );
        mlt_properties_set_position( filter_props, "_last_position", current_position );

        // Ramp from the previous gain to the current
        gain = previous_gain;

        int16_t *p = *buffer;

        // Apply the gain
        for ( i = 0; i < *samples; i++ )
        {
                for ( j = 0; j < *channels; j++ )
                {
                        sample = *p * gain;
                        *p = ROUND( sample );
                
                        if ( gain > 1.0 )
                        {
                                /* use limiter function instead of clipping */
                                if ( normalise )
                                        *p = ROUND( samplemax * limiter( sample / (double) samplemax, limiter_level ) );
                                
                                /* perform clipping */
                                else if ( sample > samplemax )
                                        *p = samplemax;
                                else if ( sample < samplemin )
                                        *p = samplemin;
                        }
                        p++;
                }
                gain += gain_step;
        }
        
        return 0;
}

/** Filter processing.
*/

static mlt_frame filter_process( mlt_filter this, mlt_frame frame )
{
        mlt_properties properties = MLT_FRAME_PROPERTIES( frame );
        mlt_properties filter_props = MLT_FILTER_PROPERTIES( this );

        // Parse the gain property
        if ( mlt_properties_get( properties, "gain" ) == NULL )
        {
                double gain = 1.0; // no adjustment
                
                if ( mlt_properties_get( filter_props, "gain" ) != NULL )
                {
                        char *p = mlt_properties_get( filter_props, "gain" );
                        
                        if ( strncaseeq( p, "normalise", 9 ) )
                                mlt_properties_set( filter_props, "normalise", "" );
                        else
                        {
                                if ( strcmp( p, "" ) != 0 )
                                        gain = fabs( strtod( p, &p) );

                                while ( isspace( *p ) )
                                        p++;

                                /* check if "dB" is given after number */
                                if ( strncaseeq( p, "db", 2 ) )
                                        gain = DBFSTOAMP( gain );
                                        
                                // If there is an end adjust gain to the range
                                if ( mlt_properties_get( filter_props, "end" ) != NULL )
                                {
                                        // Determine the time position of this frame in the transition duration
                                        mlt_position in = mlt_filter_get_in( this );
                                        mlt_position out = mlt_filter_get_out( this );
                                        mlt_position time = mlt_frame_get_position( frame );
                                        double position = ( double )( time - in ) / ( double )( out - in + 1 );

                                        double end = -1;
                                        char *p = mlt_properties_get( filter_props, "end" );
                                        if ( strcmp( p, "" ) != 0 )
                                                end = fabs( strtod( p, &p) );

                                        while ( isspace( *p ) )
                                                p++;

                                        /* check if "dB" is given after number */
                                        if ( strncaseeq( p, "db", 2 ) )
                                                end = DBFSTOAMP( gain );

                                        if ( end != -1 )
                                                gain += ( end - gain ) * position;
                                }
                        }
                }
                mlt_properties_set_double( properties, "volume.gain", gain );
        }
        
        // Parse the maximum gain property
        if ( mlt_properties_get( filter_props, "max_gain" ) != NULL )
        {
                char *p = mlt_properties_get( filter_props, "max_gain" );
                double gain = fabs( strtod( p, &p) ); // 0 = no max
                        
                while ( isspace( *p ) )
                        p++;

                /* check if "dB" is given after number */
                if ( strncaseeq( p, "db", 2 ) )
                        gain = DBFSTOAMP( gain );
                        
                mlt_properties_set_double( properties, "volume.max_gain", gain );
        }

        // Parse the limiter property
        if ( mlt_properties_get( filter_props, "limiter" ) != NULL )
        {
                char *p = mlt_properties_get( filter_props, "limiter" );
                double level = 0.5; /* -6dBFS */ 
                if ( strcmp( p, "" ) != 0 )
                        level = strtod( p, &p);
                
                while ( isspace( *p ) )
                        p++;
                
                /* check if "dB" is given after number */
                if ( strncaseeq( p, "db", 2 ) )
                {
                        if ( level > 0 )
                                level = -level;
                        level = DBFSTOAMP( level );
                }
                else
                {
                        if ( level < 0 )
                                level = -level;
                }
                mlt_properties_set_double( properties, "volume.limiter", level );
        }

        // Parse the normalise property
        if ( mlt_properties_get( filter_props, "normalise" ) != NULL )
        {
                char *p = mlt_properties_get( filter_props, "normalise" );
                double amplitude = 0.2511886431509580; /* -12dBFS */
                if ( strcmp( p, "" ) != 0 )
                        amplitude = strtod( p, &p);

                while ( isspace( *p ) )
                        p++;

                /* check if "dB" is given after number */
                if ( strncaseeq( p, "db", 2 ) )
                {
                        if ( amplitude > 0 )
                                amplitude = -amplitude;
                        amplitude = DBFSTOAMP( amplitude );
                }
                else
                {
                        if ( amplitude < 0 )
                                amplitude = -amplitude;
                        if ( amplitude > 1.0 )
                                amplitude = 1.0;
                }
                
                // If there is an end adjust gain to the range
                if ( mlt_properties_get( filter_props, "end" ) != NULL )
                {
                        // Determine the time position of this frame in the transition duration
                        mlt_position in = mlt_filter_get_in( this );
                        mlt_position out = mlt_filter_get_out( this );
                        mlt_position time = mlt_frame_get_position( frame );
                        double position = ( double )( time - in ) / ( double )( out - in + 1 );
                        amplitude *= position;
                }
                mlt_properties_set_int( properties, "volume.normalise", 1 );
                mlt_properties_set_double( properties, "volume.amplitude", amplitude );
        }

        // Parse the window property and allocate smoothing buffer if needed
        int window = mlt_properties_get_int( filter_props, "window" );
        if ( mlt_properties_get( filter_props, "smooth_buffer" ) == NULL && window > 1 )
        {
                // Create a smoothing buffer for the calculated "max power" of frame of audio used in normalisation
                double *smooth_buffer = (double*) calloc( window, sizeof( double ) );
                int i;
                for ( i = 0; i < window; i++ )
                        smooth_buffer[ i ] = -1.0;
                mlt_properties_set_data( filter_props, "smooth_buffer", smooth_buffer, 0, free, NULL );
        }
        
        // Put a filter reference onto the frame
        mlt_properties_set_data( properties, "filter_volume", this, 0, NULL, NULL );

        // Override the get_audio method
        mlt_frame_push_audio( frame, filter_get_audio );

        return frame;
}

/** Constructor for the filter.
*/

mlt_filter filter_volume_init( mlt_profile profile, mlt_service_type type, const char *id, char *arg )
{
        mlt_filter this = calloc( sizeof( struct mlt_filter_s ), 1 );
        if ( this != NULL && mlt_filter_init( this, NULL ) == 0 )
        {
                mlt_properties properties = MLT_FILTER_PROPERTIES( this );
                this->process = filter_process;
                if ( arg != NULL )
                        mlt_properties_set( properties, "gain", arg );

                mlt_properties_set_int( properties, "window", 75 );
                mlt_properties_set( properties, "max_gain", "20dB" );
        }
        return this;
}