1 files changed, 83 insertions, 0 deletions
diff --git a/SampleBin.mdwn b/SampleBin.mdwn
new file mode 100644
index 0000000..13c2a17
--- /dev/null
+++ b/SampleBin.mdwn
@@ -0,0 +1,83 @@
+
+
+# SampleBin
+
+Event sounds and games need a convinience api to trigger short sound snippets. 
+
+
+## Concepts
+
+
+### Caching
+
+Event sounds and in-game sounds are usually relaive short audio clips (<2sec). Having them predecoded would allow to trigger them more or less immediately. We could do that by getting a filedescriptor from fileno(tmpfile()) / mkstemp("...XXXXXX") and using that with: 
+[[!format txt """
+gnomevfssrc ! decodebin ! capsfilter ! fdsink sync=false
+"""]]
+When playing, we would use: 
+[[!format txt """
+fdsrc ! capsfilter ! volume ! audiopanorama ! adder
+"""]]
+For uncached sounds we can use: 
+[[!format txt """
+gnomevfssrc ! decodebin ! volume ! audiopanorama ! adder
+"""]]
+When doing the cache on demand, we can do: 
+[[!format txt """
+gnomevfssrc ! decodebin ! tee name=t ! queue ! capsfilter ! fdsink sync=false
+t. queue ! volume ! audiopanorama ! adder
+"""]]
+Each pipeline-fragment should be implemented as a bin. 
+
+
+### Mixing
+
+Mixing is done with the adder. It will have one static conection - the background music or silence. This keeps the stream continous. 
+
+When setting up the mixer we need to setup n playback subpipelines. We check if there are samples for each caching mode. Then we create n pipelines of each type (voice slots), if there are samples of that type (We need to create caching ones if there are cache after use ones). For each subpipeline we add a event-probe to the src pad of the audiopanorama (the proxy-pad of the bin in fact). This will check for eos in that bin and once it gets eos blocks the pad and marks the bin as free.  
+
+When playing a sound, we need to check for a free voice slot. If we found none, we could: * kill the one that is finished next * kill the one that is the most silent * fail * ... 
+
+For good performance we should add gap handling to the adder. 
+
+
+## API
+
+
+[[!format txt """
+gchar* !GstSampleBin::background-music-uri, rw, default=NULL
+// Set an uri to a media file that will be played in a loop. If %NULL silence is used.
+
+gdouble !GstSampleBin::background-music-volume, rw, default=1.0
+// Volume level for background music, unused if there is no background music
+
+
+typedef enum {
+  GST_SAMPLE_BIN_CACHE_NEVER,
+  GST_SAMPLE_BIN_CACHE_ALWAYS,
+  GST_SAMPLE_BIN_CACHE_AFTER_USE
+} !GstSampleBinCacheMode;
+
+glong gst_sample_bin_add_sound(!GstsampleBin *sbin,const gchar *uri, !GstSampleBinCacheMode cache_mode);
+// Set a sound sample clip and eventually preload. Returns a clip-id.
+
+glong gst_sample_bin_remove_sound(!GstsampleBin *sbin, glong clip_id);
+// Remove a sound-clip. All clips are removed when destroying.
+
+gst_sample_bin_play_sound(!GstsampleBin *sbin, glong clip_id, gdouble volume, gdouble pan);
+// Play the sound clip with the given volume and panorama position.
+"""]]
+
+# Comments
+
+* can we consider mixing as a separate functionnality? 
+* background could be considered as a regular sample 
+* a sample as a (mini)object, shareable in different contexts (refcount), and with attached (dynamic?) properties (vol,pan?,loop,name..)   
+* add and remove as part of ref/unref and internal sample/memory management 
+* playsound as a method of a sample object, with more state control?: start/stop (as props?) 
+-- [[MarcAndreLureau|MarcAndreLureau]] 2008-02-12 08:47:20 
+
+* why mixing as a separate components? it will use adder 
+* the rationale for background is, that we need something to keep adder running, and background would automatically loop on eos (segmented seek) 
+* I want volume and panorama be parameters of play action. then e.g. you can play the same sound at different volumes or pan-positions 
+-- [[StefanKost|StefanKost]] 2008-02-12 09:20:43