From Javier Taibo,
"I have found a "bug" in the new audio decoding code (actually I think the bug is in ffmpeg, but anyway it should be wise to protect the OSG plug-in about it). I am attaching a security check in FFmpegDecoderAudio.cpp.
If anybody is curious about the problem, it happens sometimes when decoding an AAC audio stream. It eventually includes a PCE block inside the AAC audio frame and then ffmpeg audio decoding function signals a "new_frame" with 1024 samples, but a null pointer instead of the audio data. It can be easily detected because in these cases number of channels is 0. Maybe this is the intended behaviour for ffmpeg, but I find it quite weird.
" It seems that libav does not have a channels attribute in AVFrame structure. This new version should do."