We are transmitting CELT encoded packages over a lossy channel, where from time to time complete packages do not arrive at the decoder. My questions: 1. What will be the best pattern to be fed into the decoder in order to minimize the audible impact and make best use of internal states from last block (like overlapp-add history etc.) and best blending with next block ? 2. Does this pattern depend on sampling frequency, bitrate, etc.? Currently we use 48kHz stereo @ 152kbit/s, 5.00ms packet size.
Markus Haist <celt at platax.de> a ?crit?:> What will be the best pattern to be fed into the decoder in order to > minimize the audible impact and make best use of internal states > from last block (like overlapp-add history etc.) and best blending > with next block ?Can you rephrase your question? The way you phrased it above, the only thing I can answer is that it's best to give to the decoder the output of the encoder. Jean-Marc
On Fri, Nov 5, 2010 at 12:31 PM, Markus Haist <celt at platax.de> wrote:> > We are transmitting CELT encoded packages over a lossy channel, where from time to time complete packages do not arrive at the decoder. > > My questions: > 1. > What will be the best pattern to be fed into the decoder in order to minimize the audible impact and make best use of internal states from last block (like overlapp-add history etc.) and best blending with next block ? > > 2. > Does this pattern depend on sampling frequency, bitrate, etc.? > Currently we use 48kHz stereo @ 152kbit/s, 5.00ms packet size.If you have lost a frame you can call the decoder with a null pointer for the input and it will do its best to conceal the loss using the internal packet loss concealment. It will correctly handle the overlap and internal state modifications on its own.