Hallo,

it was discussed on irc, but i will make it us bug report too, to make it formal.
By starting gstreamer pipeline audioresample and audioconvert prefer to set other format to pulsesrc, different from what encoder (speexenc) need.

research with "perf record empathy-av" show high load by audioresample (30%) what is higher than speexenc (17%) need.


To workaround this problem, i set capsfilter direct after pulsesrc.
This make audioresample sit silent.

Here is my workaround patch. The target is to fix gstreamer.


diff --git a/libempathy-gtk/empathy-audio-src.c b/libempathy-gtk/empathy-audio-src.c
index a3416f2..516fa27 100644
--- a/libempathy-gtk/empathy-audio-src.c
+++ b/libempathy-gtk/empathy-audio-src.c
@@ -92,6 +92,8 @@ empathy_audio_src_init (EmpathyGstAudioSrc *obj)
 {
   EmpathyGstAudioSrcPrivate *priv = EMPATHY_GST_AUDIO_SRC_GET_PRIVATE (obj);
   GstPad *ghost, *src;
+  GstElement *capsfilter;
+  GstCaps *caps;

   priv->peak_level = -G_MAXDOUBLE;
   priv->lock = g_mutex_new ();
@@ -105,11 +107,22 @@ empathy_audio_src_init (EmpathyGstAudioSrc *obj)

   fs_element_added_notifier_add (priv->notifier, GST_BIN (priv->src));

+  capsfilter = gst_element_factory_make ("capsfilter", NULL);
+  caps = gst_caps_new_simple ("audio/x-raw-int",
+    "rate", G_TYPE_INT, 8000,
+    "width", G_TYPE_INT, 16,
+    "depth", G_TYPE_INT, 16,
+    NULL);
+
+  g_object_set (G_OBJECT (capsfilter), "caps", caps, NULL);
+  gst_bin_add (GST_BIN (obj), capsfilter);
+  gst_element_link (priv->src, capsfilter);
+
   priv->volume = gst_element_factory_make ("volume", NULL);
   g_object_ref (priv->volume);

   gst_bin_add (GST_BIN (obj), priv->volume);
-  gst_element_link (priv->src, priv->volume);
+  gst_element_link (capsfilter, priv->volume);

   priv->level = gst_element_factory_make ("level", NULL);
   gst_bin_add (GST_BIN (obj), priv->level);