[opus] Opus for ASR - update and questions

Ross Bencina rossb-lists at audiomulch.com
Fri Dec 7 15:37:28 PST 2012

On 8/12/2012 10:01 AM, Young, Milan wrote:
> I'm assuming that by file pattern, you are referring to Ross' method
> of capturing actual lossey transmissions, and mapping those errors
> onto the decoded waveform.

If it's in a file it can be anything. A synthetic loss sequence or a
sampled one. Make it by hand. Write an R script. Whatever.

 > Unquantified - What loss and delay level are we simulating?  It needs
 > to have a name so it can be compared to other levels.  You can map
 > from a name into a set of error patterns, but you can't map from an
 > error pattern into a single name.

It's a little unclear here whether you're suggesting that Xiph should 
include an advanced packet loss model in the test app or not. If so, 
then you get into questions of which one, and why that one.

Personally I think the data file is an excellent way to handle it. Then 
any model can be used.

 > Non trivial - Maybe I'm behind the times, but serializing loss and
 > delays in live traffic sounds difficult.  This problem is compounded
 > by the relative infrequency of loss events.  (Most of what I'm
 > interested in are worst case scenarios north of 5% errors).

It's easy to capture an RTP stream with wireshark and run a query to 
spit out the RTP sequence numbers. From there deducing the loss pattern 
is a short script away.

 > Unrepresentative - We need a variety of error patterns at each error
 > level to ensure statistically significant results.  For my large
 > scale measurements, were talking in the thousands.

It's not too hard to go and capture thousands of packet streams... 
whether they will be representative depends on how much network access 
you have. You could easily sit on a university LAN segment and slurp up 
all the RTP traffic. But unless you do that at 1000 universities you 
will not have very good coverage -- and even then, you're only looking 
at univeristy traffic. You could do the same thing at Starbucks.*

So I agree, it's unlikely to be representative. But neither is a model 
unless it has been based on such a survey.


* But make sure you consult your lawyer first.

More information about the opus mailing list