opus - Jun 2014 - opus_multistream_encode_float not working in libopus 1.1

On Wed, Jun 4, 2014 at 4:31 PM, Timothy B. Terriberry <tterribe at
xiph.org> wrote:
> Alpha Thinktink wrote:
>> max_data_bytes=-11
>
> That value is passed in by you. I also don't think passing such a value
> would have worked in earlier releases. It indicates the size of the
> buffer you are passing to the encoder to receive the encoded output.
Actually, I'm passing a massive buffer size.  The debugger shows I'm
passing 10200 bytes, which lines up with my source code:

                while(RemoveFloatSamples(ack,packetsamplescount))
                {
                        char *tout=new char[1275*2*Fnch];
                        int
len=encode_float(e_state,ack,packetframescount,tout,1275*2*Fnch);
                        if(len>0)
                        {

AddOpusPacket(tout,len,packetframescount*rscale);
                        }
                        else
                        {
                delete [] tout;
                delete [] ack;
                Ferror=-1;
                return 0;
                        }
                }

On Wed, Jun 4, 2014 at 4:41 PM, Alpha Thinktink <alphathinktink at
gmail.com>
wrote:
>
> >> max_data_bytes=-11
>
This can happen with multistream cbr if the bitrate is set too low, to a
value that would allow less than 4 bytes per stream, per packet.  What
bitrate are you using?  Perhaps you set it to a value in kb/s instead of
b/s.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.xiph.org/pipermail/opus/attachments/20140604/d334eff6/attachment.htm

On Wed, Jun 4, 2014 at 5:08 PM, Mark Harris <mark.hsj at gmail.com>
wrote:
> On Wed, Jun 4, 2014 at 4:41 PM, Alpha Thinktink <alphathinktink at
gmail.com>
> wrote:
>>
>> >> max_data_bytes=-11
>
>
> This can happen with multistream cbr if the bitrate is set too low, to a
> value that would allow less than 4 bytes per stream, per packet.  What
> bitrate are you using?  Perhaps you set it to a value in kb/s instead of
> b/s.
I just verified I'm not sending in kb/s.  The user sets the value as
either Auto (-1000), Max (-1), or Specify --> between 512000-500  The
value is passed, as-is, to the encoder ctl function.

However, it looks like it's getting closer:
Quadraphonic signal: 4800b/s=go  4799b/s=no-go, so anything below 4800
doesn't work.
Stereo signal (as 2chan uncoupled):  1600b/s=go  1599b/s=no-go, here
nothing below 1600 works.

However, before testing different bitrates I've had it set on Auto,
which is no-go on v1.1 Testing Max also is no-go v1.1  However,
v1.1-beta and earlier works.

Please remember that everything still works with v1.1-beta and
previous.  I just tested it to make sure as well.  v1.1 release is the
only version I see problems with hard cbr multistream.

In debugging I saw:
opus_multistream_encoder.c
line 760
if(!vbr)
          max_data_bytes = IMIN(max_data_bytes,
3*st->bitrate_bps/(3*8*Fs/frame_size));

max_data_bytes after this code becomes -2 where as before it was
10200.  I suspect it was because st->bitrate_bps was set to -1000 to
indicate Auto bitrate.  I imagine Max bitrate (-1) has a slightly
similar effect.

then in the same c file
line 846
        curr_max = max_data_bytes - tot_size;

curr_max becomes -2 as tot_size was 0.

then line 848

        curr_max -= IMAX(0,4*(st->layout.nb_streams-s-1)-1);

cur_max becomes -13 as nb_streams was 4 and s was 0.

then line 849

        curr_max = IMIN(curr_max,MS_FRAME_TMP);

which left curr_max at -13

The result at line 853 is opus_encode_native is called with parameters
so thusly:
(encoder state pointer, buf pointer, frame_size=960, tmp_data pointer,
curr_max=-13, lsb_depth=24, pcm pointer, analysis_frame_size=960,
c1=0, c2=-1, st->layout.nb_channels=4, downmix pointer, float_api=1);


I don't know what to suggest as I've never tinkered with the opus
library internals, nor have I ever compiled it before today and do not
know what it looked like in previous versions.  Sorry.

Hi,

I just checked in a fix for the bug you reported. Let me know if it works.

Cheers,

	Jean-Marc

On 05/06/14 12:41 AM, Alpha Thinktink wrote:
> In debugging I saw:
> opus_multistream_encoder.c
> line 760
> if(!vbr)
>           max_data_bytes = IMIN(max_data_bytes,
> 3*st->bitrate_bps/(3*8*Fs/frame_size));
> 
> max_data_bytes after this code becomes -2 where as before it was
> 10200.  I suspect it was because st->bitrate_bps was set to -1000 to
> indicate Auto bitrate.  I imagine Max bitrate (-1) has a slightly
> similar effect.
> 
> then in the same c file
> line 846
>         curr_max = max_data_bytes - tot_size;
> 
> curr_max becomes -2 as tot_size was 0.
> 
> then line 848
> 
>         curr_max -= IMAX(0,4*(st->layout.nb_streams-s-1)-1);
> 
> cur_max becomes -13 as nb_streams was 4 and s was 0.
> 
> then line 849
> 
>         curr_max = IMIN(curr_max,MS_FRAME_TMP);
> 
> which left curr_max at -13
> 
> The result at line 853 is opus_encode_native is called with parameters
> so thusly:
> (encoder state pointer, buf pointer, frame_size=960, tmp_data pointer,
> curr_max=-13, lsb_depth=24, pcm pointer, analysis_frame_size=960,
> c1=0, c2=-1, st->layout.nb_channels=4, downmix pointer, float_api=1);
> 
> 
> I don't know what to suggest as I've never tinkered with the opus
> library internals, nor have I ever compiled it before today and do not
> know what it looked like in previous versions.  Sorry.
> _______________________________________________
> opus mailing list
> opus at xiph.org
> http://lists.xiph.org/mailman/listinfo/opus
>