Started with VAD

toxav/Makefile.inc

@@ -10,8 +10,8 @@ libtoxav_la_SOURCES = ../toxav/event.h \
                       ../toxav/rtp.c \
                       ../toxav/msi.h \
                       ../toxav/msi.c \
-                      ../toxav/media.h \
-                      ../toxav/media.c \
+                      ../toxav/codec.h \
+                      ../toxav/codec.c \
                       ../toxav/toxav.h \
                       ../toxav/toxav.c
 

toxav/codec.c

@@ -34,7 +34,8 @@
 #include <assert.h>
 
 #include "rtp.h"
-#include "media.h"
+#include "codec.h"
+
 
 int empty_queue(JitterBuffer *q)
 {
@@ -292,7 +293,10 @@ CodecState *codec_init_session ( uint32_t audio_bitrate,
         free (retu);
         return NULL;
     }
-
+    
+    float frame_duration_sec = audio_frame_duration / 1000;
+    retu->samples_per_frame = audio_sample_rate * frame_duration_sec;    
+    
     return retu;
 }
 
@@ -314,3 +318,28 @@ void codec_terminate_session ( CodecState *cs )
     if ( cs->capabilities & v_encoding )
         vpx_codec_destroy(&cs->v_encoder);
 }
+
+inline float calculate_sum_sq (int16_t* n, size_t k)
+{
+    float result = 0;
+    size_t i = 0;
+    
+    for ( ; i < k; i ++) {
+        result += (float) (n[i] * n[i]);
+    }
+    
+    return result;
+}
+
+int calculate_VAD_from_PCM( int16_t* PCM, size_t frame_size, float energy)
+{
+//     int i = 0;
+//     for (; i < frame_size; i ++) {
+    LOGGER_DEBUG("Frame size: %d ref: %f", frame_size, energy);
+    float frame_energy = sqrt(calculate_sum_sq(PCM, frame_size)) / frame_size;
+    LOGGER_DEBUG("Frame energy calculated: %f", frame_energy);
+    if ( frame_energy > energy) return 1;
+//     }
+
+    return 0;
+}

toxav/codec.h

@@ -21,8 +21,8 @@
  *
  */
 
-#ifndef _AVCODEC_H_
-#define _AVCODEC_H_
+#ifndef _CODEC_H_
+#define _CODEC_H_
 
 #include <stdio.h>
 #include <math.h>
@@ -46,6 +46,8 @@ typedef enum _Capabilities {
     v_decoding = 1 << 3
 } Capabilities;
 
+
+
 typedef struct _CodecState {
 
     /* video encoding */
@@ -64,7 +66,9 @@ typedef struct _CodecState {
     OpusDecoder *audio_decoder;
 
     uint64_t capabilities; /* supports*/
-
+    
+    /* Voice activity detection */
+    float samples_per_frame;
 } CodecState;
 
 
@@ -92,8 +96,12 @@ CodecState *codec_init_session ( uint32_t audio_bitrate,
                                  uint32_t audio_channels,
                                  uint16_t video_width,
                                  uint16_t video_height,
-                                 uint32_t video_bitrate );
+                                 uint32_t video_bitrate);
 
 void codec_terminate_session(CodecState *cs);
 
-#endif
+
+/* return 1 if has voice, 0 if not */
+int calculate_VAD_from_PCM(int16_t* PCM, size_t frame_size, float energy);
+
+#endif /* _CODEC_H_ */

toxav/toxav.c

@@ -27,13 +27,12 @@
 #define _GNU_SOURCE /* implicit declaration warning */
 
 #include "rtp.h"
-#include "media.h"
+#include "codec.h"
 #include "msi.h"
 #include "toxav.h"
 
 #include "../toxcore/logger.h"
 
-
 #include <assert.h>
 #include <stdlib.h>
 #include <string.h>
@@ -732,4 +731,9 @@ inline__ int toxav_set_video_queue_limit(ToxAv *av, int32_t call_index, uint64_t
 inline__ Tox *toxav_get_tox(ToxAv *av)
 {
     return (Tox *)av->messenger;
+}
+
+int toxav_has_activity(int16_t* PCM, uint16_t frame_size, float ref_energy)
+{
+    return calculate_VAD_from_PCM(PCM, frame_size, ref_energy);
 }

toxav/toxav.h

@@ -90,7 +90,8 @@ typedef enum {
     ErrorTerminatingAudioRtp = -9, /* Returned in toxav_kill_transmission() */
     ErrorTerminatingVideoRtp = -10, /* Returned in toxav_kill_transmission() */
     ErrorPacketTooLarge = -11, /* Buffer exceeds size while encoding */
-
+    ErrorInvalidCodecState = -12, /* Codec state not initialized */
+    
 } ToxAvError;
 
 
@@ -373,6 +374,8 @@ int toxav_set_video_queue_limit ( ToxAv *av, int32_t call_index, uint64_t limit
 
 Tox *toxav_get_tox(ToxAv *av);
 
+int toxav_has_activity ( int16_t* PCM, uint16_t frame_size, float ref_energy );
+
 #ifdef __cplusplus
 }
 #endif