[asterisk-bugs] [JIRA] (ASTERISK-27610) app_amd.so returning TOOLONG before reaching the timeout

[Richard Mudgett (JIRA)]

    [ https://issues.asterisk.org/jira/browse/ASTERISK-27610?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=242068#comment-242068 ] 

Richard Mudgett commented on ASTERISK-27610:
--------------------------------------------

For AMD that is an acceptable way to calculate the frame length in ms because the read format is always going to be slin.  The AMD application sets the channel's read format to ast_format_slin before going into the ast_waitfor() loop so the DSP code can process the audio signal.

> app_amd.so returning TOOLONG before reaching the timeout
> --------------------------------------------------------
>
>                 Key: ASTERISK-27610
>                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-27610
>             Project: Asterisk
>          Issue Type: Bug
>      Security Level: None
>          Components: Applications/app_amd
>    Affects Versions: 13.17.2
>         Environment: Vicidial dial server running OpenSuSE 42.1 on Intel hardware
>            Reporter: Michael Cargile
>            Assignee: Unassigned
>              Labels: patch
>         Attachments: agi_explainations.txt, amd.patch, cli.txt, dialplan.txt, manager_events.txt
>
>
> Calls that pass through AMD are returning TOOLONG about 70% of the time even though they have not reached the analysis timeout. Here is an example:
> {noformat}
> [Jan 18 12:44:12]     -- Executing [8369 at default:2] AMD("SIP/sgw2-00000007", "2000,2000,1000,5000,120,50,4,256") in new stack
> [Jan 18 12:44:12]     -- AMD: SIP/sgw2-00000007 (N/A) (N/A) (Fmt: slin)
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Changed state to STATE_IN_SILENCE
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 40
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 40
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 100
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 100
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 100
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 100
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 100
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Detected Talk, previous silence duration: 200
> [Jan 18 12:44:12]     -- AMD: Channel [SIP/sgw2-00000007]. Short Word Duration: 20
> [Jan 18 12:44:13]     -- AMD: Channel [SIP/sgw2-00000007]. Too long...
> {noformat}
> TotalAnalysis is set to 5000 or 5 seconds and yet according to the CLI at most 2 seconds has elapsed since the AMD app was initiated.
> I ran tests on both Asterisk 11 and Asterisk 13 comparing the results. I had two Vicidial servers setup. One with 11 the other with 13. I had them calling into two different IVRs. The first IVR played an audio file that was just saying "Hello". The second IVR played an audio file that sounded like your typical answering machine message, "Hello. You have reached the blah residence. Please leave a messsage." Neither audio file was over 3 seconds long. I then called each number 50k times from both dialers. So each dialer placed a total of 100k calls. Asterisk 11 only had 123 TOOLONG messages most likely from poor audio connection. Asterisk 13 on the other hand had 70k+ TOOLONGs.
> Also the results for the non-TOOLONG calls on Asterisk 13 were very poor. It only seems to be about 50% accurate at determining if the audio is the answering machine message or just "hello". My guess is that it is basically randomly selecting one, but that is hard to say.
> From what I can tell from the code not much has changed in app_amd from Asterisk 11 to Asterisk 13. My guess is some core resource it relies on for timing has changed and it was not adjusted accordingly.
> I will attached details on how to reproduce this in the comments.



--
This message was sent by Atlassian JIRA
(v6.2#6252)