opus - Oct 2013 - Regarding error handling in Opus Decoder

Hi,

While testing the Opus Decoder we came across the following:
If the decoder encounters an 'Invalid Payload Length' the decoding of
the stream is stopped.
Also, when the decoder encounters 'Range coder state mismatch', the
decoding of the stream is stopped.
I believe that it should reject the erroneous payload and start decoding the
next payload for it to work properly in a streaming environment.

Is this an expected behavior?

Regards,
Rhishi

-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/opus/attachments/20131004/ee577d1c/attachment.htm

On 2013-10-04 1:34 PM, Rhishikesh Agashe wrote:
> I believe that it should reject the erroneous payload and start decoding
> the next payload for it to work properly in a streaming environment.
Because the decoder's predictor state converges over multiple packets of
data, you cannot just skip to the next payload without introducing
decoding errors.

We have also tried to make the reference implementation strict to help
dectect data corruption bugs during deployment.

Thus, stopping the decoder on corrupt data is the expected behaviour.

How are you encountering these conditions?

'Invalid Payload Length' and especially 'Range coder state
mismatch'
indicate either corrupt data for a decoder bug. Corrupt packets should
be detected and discarded by the network layer, in which case you can
use the packet loss concealment and/or forward error correction support
to minimize the damage.

 -r

Hi Ralph,

We tried to test the Opus decoder using the attached vectors and got the below
mentioned  errors.
If the errors are to be detected at the network layer, then, I believe that the
conditions in these vectors won't occur till the control reaches the opus
decoder.

Do let me know if I have understood it correctly.

Thanks and Regards,
Rhishi


-----Original Message-----
From: Ralph Giles [mailto:giles at thaumas.net] 
Sent: Sunday, October 06, 2013 11:51
To: Rhishikesh Agashe; opus at xiph.org
Cc: Rasmi Mishra
Subject: Re: [opus] Regarding error handling in Opus Decoder

On 2013-10-04 1:34 PM, Rhishikesh Agashe wrote:
> I believe that it should reject the erroneous payload and start 
> decoding the next payload for it to work properly in a streaming
environment.
Because the decoder's predictor state converges over multiple packets of
data, you cannot just skip to the next payload without introducing decoding
errors.

We have also tried to make the reference implementation strict to help dectect
data corruption bugs during deployment.

Thus, stopping the decoder on corrupt data is the expected behaviour.

How are you encountering these conditions?

'Invalid Payload Length' and especially 'Range coder state
mismatch'
indicate either corrupt data for a decoder bug. Corrupt packets should be
detected and discarded by the network layer, in which case you can use the
packet loss concealment and/or forward error correction support to minimize the
damage.

 -r
-------------- next part --------------
A non-text attachment was scrubbed...
Name: testvector12_corrupt.bit
Type: application/octet-stream
Size: 56702 bytes
Desc: testvector12_corrupt.bit
Url :
http://lists.xiph.org/pipermail/opus/attachments/20131008/3436d64e/attachment-0002.obj
-------------- next part --------------
A non-text attachment was scrubbed...
Name: testvector08_corrupt.bit
Type: application/octet-stream
Size: 140061 bytes
Desc: testvector08_corrupt.bit
Url :
http://lists.xiph.org/pipermail/opus/attachments/20131008/3436d64e/attachment-0003.obj