[opus] Strange behavior with OGG packets?

[Mark Harris]

On Thu, May 18, 2017 at 12:23 PM, Drew Allen <bitllama at google.com> wrote:
> I should also say the the ogg_packet.packet[] char array is zero-filled for
> packet mentioned.
>
> On Thu, May 18, 2017 at 12:19 PM Drew Allen <bitllama at google.com> wrote:
>>
>> Hello all,
>>
>> I'm currently in the process of implementing a mixing matrix into the
>> header of OPUS files for the proposed channel mapping 3 seen here:
>> https://tools.ietf.org/html/draft-ietf-codec-ambisonics-03#section-3.2
>>
>> My implementation simply replaces writing the stream mapping char array
>> with the Q15 col-wise matrix vals described in the above document.
>>
>> I'm able to write/read matrices of 4x4 and 9x9 and encode->decode first
>> and second order just fine, but writing larger matrices causes what seems to
>> be some sort of packet corruption. The OpusHead packet is never found, but
>> in it's place we have a packet at b_o_s=256 (which is where OpusHead usually
>> lives), but the byte size is reported to be 0 and e_o_s=0 as well. (This is
>> generated by parsing the packets via ogg's ogg_stream_packetout function. It
>> appears a matrix larger than 7x16 causes this issue. (The matrix values are
>> all signed short vals, so >224 bytes seems to fail.
>>
>> I'm not very familiar with ogg internals so any advice would be greatly
>> appreciated! Thanks so much! :)
>>
>> Cheers,
>> Drew


Hi Drew,

The packet is probably larger than the buffer.  In opus-tools, see
src/opusenc.c lines 849-853.

 - Mark