Resample audio channels when doing real stereo to mono.

modules/audio_filter/converter/mono.c

@@ -26,6 +26,7 @@
  *****************************************************************************/
 #include <stdlib.h>                                      /* malloc(), free() */
 #include <string.h>
+#include <math.h>                                        /* sqrt */
 
 #ifdef HAVE_STDINT_H
 #   include <stdint.h>                                         /* int16_t .. */
@@ -54,19 +55,41 @@ static block_t *Convert( filter_t *p_filter, block_t *p_block );
 
 static unsigned int stereo_to_mono( aout_instance_t *, aout_filter_t *,
                                     aout_buffer_t *, aout_buffer_t * );
+static void stereo2mono_downmix( aout_instance_t *, aout_filter_t *,
+                                 aout_buffer_t *, aout_buffer_t * );
 
 /*****************************************************************************
  * Local structures
  *****************************************************************************/
+struct atomic_operation_t
+{
+    int i_source_channel_offset;
+    int i_dest_channel_offset;
+    unsigned int i_delay;/* in sample unit */
+    double d_amplitude_factor;
+};
+
 struct filter_sys_t
 {
-    int i_nb_channels; /* number of float32 per sample */
+    vlc_bool_t b_downmix;
+
+    int i_nb_channels; /* number of int16_t per sample */
     int i_channel_selected;
     int i_bitspersample;
+
+    size_t i_overflow_buffer_size;/* in bytes */
+    byte_t * p_overflow_buffer;
+    unsigned int i_nb_atomic_operations;
+    struct atomic_operation_t * p_atomic_operations;
 };
 
+#define MONO_DOWNMIX_TEXT ("Use downmix algorithme.")
+#define MONO_DOWNMIX_LONGTEXT ("This option selects a stereo to mono " \
+    "downmix algorithm that is used in the headphone channel mixer. It" \
+    "gives the effect of standing in a room full of speakers." )
+
 #define MONO_CHANNEL_TEXT ("Select channel to keep")
-#define MONO_CHANNEL_LONGTEXT ("This option silcences all other channels " \
+#define MONO_CHANNEL_LONGTEXT ("This option silences all other channels " \
     "except the selected channel. Choose one from (0=left, 1=right " \
     "2=rear left, 3=rear right, 4=center, 5=left front)")
 
@@ -88,7 +111,10 @@ vlc_module_begin();
     set_description( _("Audio filter for stereo to mono conversion") );
     set_capability( "audio filter2", 5 );
 
-    add_integer( MONO_CFG "mono-channel", -1, NULL, MONO_CHANNEL_TEXT, MONO_CHANNEL_LONGTEXT, VLC_FALSE );
+    add_bool( MONO_CFG "mono-downmix", VLC_FALSE, NULL, MONO_DOWNMIX_TEXT,
+              MONO_DOWNMIX_LONGTEXT, VLC_FALSE );
+    add_integer( MONO_CFG "mono-channel", -1, NULL, MONO_CHANNEL_TEXT,
+        MONO_CHANNEL_LONGTEXT, VLC_FALSE );
         change_integer_list( pi_pos_values, ppsz_pos_descriptions, 0 );
 
     set_category( CAT_AUDIO );
@@ -97,6 +123,238 @@ vlc_module_begin();
     set_shortname( "Mono" );
 vlc_module_end();
 
+/* Init() and ComputeChannelOperations() -
+ * Code taken from modules/audio_filter/channel_mixer/headphone.c
+ * converted from float into int16_t based downmix
+ * Written by Boris Dorès <babal@via.ecp.fr>
+ */
+
+/*****************************************************************************
+ * Init: initialize internal data structures
+ * and computes the needed atomic operations
+ *****************************************************************************/
+/* x and z represent the coordinates of the virtual speaker
+ *  relatively to the center of the listener's head, measured in meters :
+ *
+ *  left              right
+ *Z
+ *-
+ *a          head
+ *x
+ *i
+ *s
+ *  rear left    rear right
+ *
+ *          x-axis
+ *  */
+static void ComputeChannelOperations( struct filter_sys_t * p_data,
+        unsigned int i_rate, unsigned int i_next_atomic_operation,
+        int i_source_channel_offset, double d_x, double d_z,
+        double d_compensation_length, double d_channel_amplitude_factor )
+{
+    double d_c = 340; /*sound celerity (unit: m/s)*/
+    double d_compensation_delay = (d_compensation_length-0.1) / d_c * i_rate;
+
+    /* Left ear */
+    p_data->p_atomic_operations[i_next_atomic_operation]
+        .i_source_channel_offset = i_source_channel_offset;
+    p_data->p_atomic_operations[i_next_atomic_operation]
+        .i_dest_channel_offset = 0;/* left */
+    p_data->p_atomic_operations[i_next_atomic_operation]
+        .i_delay = (int)( sqrt( (-0.1-d_x)*(-0.1-d_x) + (0-d_z)*(0-d_z) )
+                          / d_c * i_rate - d_compensation_delay );
+    if( d_x < 0 )
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation]
+            .d_amplitude_factor = d_channel_amplitude_factor * 1.1 / 2;
+    }
+    else if( d_x > 0 )
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation]
+            .d_amplitude_factor = d_channel_amplitude_factor * 0.9 / 2;
+    }
+    else
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation]
+            .d_amplitude_factor = d_channel_amplitude_factor / 2;
+    }
+
+    /* Right ear */
+    p_data->p_atomic_operations[i_next_atomic_operation + 1]
+        .i_source_channel_offset = i_source_channel_offset;
+    p_data->p_atomic_operations[i_next_atomic_operation + 1]
+        .i_dest_channel_offset = 1;/* right */
+    p_data->p_atomic_operations[i_next_atomic_operation + 1]
+        .i_delay = (int)( sqrt( (0.1-d_x)*(0.1-d_x) + (0-d_z)*(0-d_z) )
+                          / d_c * i_rate - d_compensation_delay );
+    if( d_x < 0 )
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation + 1]
+            .d_amplitude_factor = d_channel_amplitude_factor * 0.9 / 2;
+    }
+    else if( d_x > 0 )
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation + 1]
+            .d_amplitude_factor = d_channel_amplitude_factor * 1.1 / 2;
+    }
+    else
+    {
+        p_data->p_atomic_operations[i_next_atomic_operation + 1]
+            .d_amplitude_factor = d_channel_amplitude_factor / 2;
+    }
+}
+
+static int Init( vlc_object_t *p_this, struct filter_sys_t * p_data,
+                 unsigned int i_nb_channels, uint32_t i_physical_channels,
+                 unsigned int i_rate )
+{
+    double d_x = config_GetInt( p_this, "headphone-dim" );
+    double d_z = d_x;
+    double d_z_rear = -d_x/3;
+    double d_min = 0;
+    unsigned int i_next_atomic_operation;
+    int i_source_channel_offset;
+    unsigned int i;
+
+    if( p_data == NULL )
+    {
+        msg_Dbg( p_this, "passing a null pointer as argument" );
+        return 0;
+    }
+
+    if( config_GetInt( p_this, "headphone-compensate" ) )
+    {
+        /* minimal distance to any speaker */
+        if( i_physical_channels & AOUT_CHAN_REARCENTER )
+        {
+            d_min = d_z_rear;
+        }
+        else
+        {
+            d_min = d_z;
+        }
+    }
+
+    /* Number of elementary operations */
+    p_data->i_nb_atomic_operations = i_nb_channels * 2;
+    if( i_physical_channels & AOUT_CHAN_CENTER )
+    {
+        p_data->i_nb_atomic_operations += 2;
+    }
+    p_data->p_atomic_operations = malloc( sizeof(struct atomic_operation_t)
+            * p_data->i_nb_atomic_operations );
+    if( p_data->p_atomic_operations == NULL )
+    {
+        msg_Err( p_this, "out of memory" );
+        return -1;
+    }
+
+    /* For each virtual speaker, computes elementary wave propagation time
+     * to each ear */
+    i_next_atomic_operation = 0;
+    i_source_channel_offset = 0;
+    if( i_physical_channels & AOUT_CHAN_LEFT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , -d_x , d_z , d_min , 2.0 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_RIGHT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , d_x , d_z , d_min , 2.0 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_MIDDLELEFT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , -d_x , 0 , d_min , 1.5 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_MIDDLERIGHT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , d_x , 0 , d_min , 1.5 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_REARLEFT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , -d_x , d_z_rear , d_min , 1.5 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_REARRIGHT )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , d_x , d_z_rear , d_min , 1.5 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_REARCENTER )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , 0 , -d_z , d_min , 1.5 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_CENTER )
+    {
+        /* having two center channels increases the spatialization effect */
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , d_x / 5.0 , d_z , d_min , 0.75 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , -d_x / 5.0 , d_z , d_min , 0.75 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+    if( i_physical_channels & AOUT_CHAN_LFE )
+    {
+        ComputeChannelOperations( p_data , i_rate
+                , i_next_atomic_operation , i_source_channel_offset
+                , 0 , d_z_rear , d_min , 5.0 / i_nb_channels );
+        i_next_atomic_operation += 2;
+        i_source_channel_offset++;
+    }
+
+    /* Initialize the overflow buffer
+     * we need it because the process induce a delay in the samples */
+    p_data->i_overflow_buffer_size = 0;
+    for( i = 0 ; i < p_data->i_nb_atomic_operations ; i++ )
+    {
+        if( p_data->i_overflow_buffer_size
+                < p_data->p_atomic_operations[i].i_delay * 2 * sizeof (int16_t) )
+        {
+            p_data->i_overflow_buffer_size
+                = p_data->p_atomic_operations[i].i_delay * 2 * sizeof (int16_t);
+        }
+    }
+    p_data->p_overflow_buffer = malloc( p_data->i_overflow_buffer_size );
+    if( p_data->p_atomic_operations == NULL )
+    {
+        msg_Err( p_this, "out of memory" );
+        return -1;
+    }
+    memset( p_data->p_overflow_buffer, 0 , p_data->i_overflow_buffer_size );
+
+    /* end */
+    return 0;
+}
+
 /*****************************************************************************
  * OpenFilter
  *****************************************************************************/
@@ -135,23 +393,45 @@ static int OpenFilter( vlc_object_t *p_this )
         return VLC_EGENERIC;
     }
 
+    var_Create( p_this, MONO_CFG "mono-downmix",
+                VLC_VAR_BOOL | VLC_VAR_DOINHERIT );
+    p_sys->b_downmix = var_GetBool( p_this, MONO_CFG "mono-downmix" );
+
     var_Create( p_this, MONO_CFG "mono-channel",
                 VLC_VAR_INTEGER | VLC_VAR_DOINHERIT );
     p_sys->i_channel_selected =
             (unsigned int) var_GetInteger( p_this, MONO_CFG "mono-channel" );
 
-#if 0
-    p_filter->fmt_out.audio.i_physical_channels = AOUT_CHAN_CENTER;
-#else
-    p_filter->fmt_out.audio.i_physical_channels =
+    if( p_sys->b_downmix )
+    {
+        p_filter->fmt_out.audio.i_physical_channels = AOUT_CHAN_CENTER;
+    }
+    else
+    {
+        p_filter->fmt_out.audio.i_physical_channels =
                             (AOUT_CHAN_LEFT | AOUT_CHAN_RIGHT);
-#endif
+    }
+    p_filter->fmt_out.audio.i_channels = aout_FormatNbChannels( &(p_filter->fmt_out.audio) );
+
     p_filter->fmt_out.audio.i_rate = p_filter->fmt_in.audio.i_rate;
     p_filter->fmt_out.audio.i_format = p_filter->fmt_out.i_codec;
 
     p_sys->i_nb_channels = aout_FormatNbChannels( &(p_filter->fmt_in.audio) );
     p_sys->i_bitspersample = p_filter->fmt_out.audio.i_bitspersample;
 
+    p_sys->i_overflow_buffer_size = 0;
+    p_sys->p_overflow_buffer = NULL;
+    p_sys->i_nb_atomic_operations = 0;
+    p_sys->p_atomic_operations = NULL;
+
+    if( Init( VLC_OBJECT(p_filter), p_filter->p_sys,
+              aout_FormatNbChannels( &p_filter->fmt_in.audio ),
+              p_filter->fmt_in.audio.i_physical_channels,
+              p_filter->fmt_in.audio.i_rate ) < 0 )
+    {
+        return VLC_EGENERIC;
+    }
+
     p_filter->pf_audio_filter = Convert;
 
     msg_Dbg( p_this, "%4.4s->%4.4s, channels %d->%d, bits per sample: %i->%i",
@@ -174,6 +454,7 @@ static void CloseFilter( vlc_object_t *p_this)
     filter_sys_t *p_sys = p_filter->p_sys;
 
     var_Destroy( p_this, MONO_CFG "mono-channel" );
+    var_Destroy( p_this, MONO_CFG "mono-downmix" );
     free( p_sys );
 }
 
@@ -185,8 +466,8 @@ static block_t *Convert( filter_t *p_filter, block_t *p_block )
     aout_filter_t aout_filter;
     aout_buffer_t in_buf, out_buf;
     block_t *p_out = NULL;
-    int i_out_size;
     unsigned int i_samples;
+    int i_out_size;
 
     if( !p_block || !p_block->i_samples )
     {
@@ -205,9 +486,8 @@ static block_t *Convert( filter_t *p_filter, block_t *p_block )
         p_block->pf_release( p_block );
         return NULL;
     }
-
     p_out->i_samples = (p_block->i_samples / p_filter->p_sys->i_nb_channels) *
-                            aout_FormatNbChannels( &(p_filter->fmt_out.audio) );
+                       aout_FormatNbChannels( &(p_filter->fmt_out.audio) );
     p_out->i_dts = p_block->i_dts;
     p_out->i_pts = p_block->i_pts;
     p_out->i_length = p_block->i_length;
@@ -236,8 +516,17 @@ static block_t *Convert( filter_t *p_filter, block_t *p_block )
     out_buf.i_nb_bytes = p_out->i_buffer;
     out_buf.i_nb_samples = p_out->i_samples;
 
-    i_samples = stereo_to_mono( (aout_instance_t *)p_filter, &aout_filter,
-                                &out_buf, &in_buf );
+    if( p_filter->p_sys->b_downmix )
+    {
+        memset( out_buf.p_buffer, 0, i_out_size );
+        stereo2mono_downmix( (aout_instance_t *)p_filter, &aout_filter,
+                             &in_buf, &out_buf );
+    }
+    else
+    {
+        i_samples = stereo_to_mono( (aout_instance_t *)p_filter, &aout_filter,
+                                    &out_buf, &in_buf );
+    }
 
     p_out->i_buffer = out_buf.i_nb_bytes;
     p_out->i_samples = out_buf.i_nb_samples;
@@ -246,9 +535,122 @@ static block_t *Convert( filter_t *p_filter, block_t *p_block )
     return p_out;
 }
 
-/* stereo_to_mono - mix 2 channels (left,right) into one and play silence on
- * all other channels.
+/* stereo2mono_downmix - stereo channels into one mono channel.
+ * Code taken from modules/audio_filter/channel_mixer/headphone.c
+ * converted from float into int16_t based downmix
+ * Written by Boris Dorès <babal@via.ecp.fr>
  */
+static void stereo2mono_downmix( aout_instance_t * p_aout, aout_filter_t * p_filter,
+                            aout_buffer_t * p_in_buf, aout_buffer_t * p_out_buf )
+{
+    filter_sys_t *p_sys = (filter_sys_t *)p_filter->p_sys;
+
+    int i_input_nb = aout_FormatNbChannels( &p_filter->input );
+    int i_output_nb = aout_FormatNbChannels( &p_filter->output );
+
+    int16_t * p_in = (int16_t*) p_in_buf->p_buffer;
+    byte_t * p_out;
+    byte_t * p_overflow;
+    byte_t * p_slide;
+
+    size_t i_overflow_size;     /* in bytes */
+    size_t i_out_size;          /* in bytes */
+
+    unsigned int i, j;
+
+    int i_source_channel_offset;
+    int i_dest_channel_offset;
+    unsigned int i_delay;
+    double d_amplitude_factor;
+
+    /* out buffer characterisitcs */
+    p_out_buf->i_nb_samples = p_in_buf->i_nb_samples;
+    p_out_buf->i_nb_bytes = p_in_buf->i_nb_bytes * i_output_nb / i_input_nb;
+    p_out = p_out_buf->p_buffer;
+    i_out_size = p_out_buf->i_nb_bytes;
+
+    if( p_sys != NULL )
+    {
+        /* Slide the overflow buffer */
+        p_overflow = p_sys->p_overflow_buffer;
+        i_overflow_size = p_sys->i_overflow_buffer_size;
+
+        memset( p_out, 0, i_out_size );
+        if ( i_out_size > i_overflow_size )
+            memcpy( p_out, p_overflow, i_overflow_size );
+        else
+            memcpy( p_out, p_overflow, i_out_size );
+
+        p_slide = p_sys->p_overflow_buffer;
+        while( p_slide < p_overflow + i_overflow_size )
+        {
+            if( p_slide + i_out_size < p_overflow + i_overflow_size )
+            {
+                memset( p_slide, 0, i_out_size );
+                if( p_slide + 2 * i_out_size < p_overflow + i_overflow_size )
+                    memcpy( p_slide, p_slide + i_out_size, i_out_size );
+                else
+                    memcpy( p_slide, p_slide + i_out_size,
+                            p_overflow + i_overflow_size - ( p_slide + i_out_size ) );
+            }
+            else
+            {
+                memset( p_slide, 0, p_overflow + i_overflow_size - p_slide );
+            }
+            p_slide += i_out_size;
+        }
+
+        /* apply the atomic operations */
+        for( i = 0; i < p_sys->i_nb_atomic_operations; i++ )
+        {
+            /* shorter variable names */
+            i_source_channel_offset
+                = p_sys->p_atomic_operations[i].i_source_channel_offset;
+            i_dest_channel_offset
+                = p_sys->p_atomic_operations[i].i_dest_channel_offset;
+            i_delay = p_sys->p_atomic_operations[i].i_delay;
+            d_amplitude_factor
+                = p_sys->p_atomic_operations[i].d_amplitude_factor;
+
+            if( p_out_buf->i_nb_samples > i_delay )
+            {
+                /* current buffer coefficients */
+                for( j = 0; j < p_out_buf->i_nb_samples - i_delay; j++ )
+                {
+                    ((int16_t*)p_out)[ (i_delay+j)*i_output_nb + i_dest_channel_offset ]
+                        += p_in[ j * i_input_nb + i_source_channel_offset ]
+                           * d_amplitude_factor;
+                }
+
+                /* overflow buffer coefficients */
+                for( j = 0; j < i_delay; j++ )
+                {
+                    ((int16_t*)p_overflow)[ j*i_output_nb + i_dest_channel_offset ]
+                        += p_in[ (p_out_buf->i_nb_samples - i_delay + j)
+                           * i_input_nb + i_source_channel_offset ]
+                           * d_amplitude_factor;
+                }
+            }
+            else
+            {
+                /* overflow buffer coefficients only */
+                for( j = 0; j < p_out_buf->i_nb_samples; j++ )
+                {
+                    ((int16_t*)p_overflow)[ (i_delay - p_out_buf->i_nb_samples + j)
+                        * i_output_nb + i_dest_channel_offset ]
+                        += p_in[ j * i_input_nb + i_source_channel_offset ]
+                           * d_amplitude_factor;
+                }
+            }
+        }
+    }
+    else
+    {
+        memset( p_out, 0, i_out_size );
+    }
+}
+
+/* Simple stereo to mono mixing. */
 static unsigned int stereo_to_mono( aout_instance_t * p_aout, aout_filter_t *p_filter,
                                     aout_buffer_t *p_output, aout_buffer_t *p_input )
 {