MediaSense Forums

« Back to Technical Discussion

RE: New Message from Daniel Hatch in Cisco MediaSense (MediaSense) - Techni

Combination View Flat View Tree View
Threads [ Previous | Next ]
I'm having a problem where one of the raw tracks provided by MediaSense is not the same length as the other one. Therefore when I mux the tracks together into a single call track people are talking over each other and it is difficult/impossible to follow the call. Because the metadata provided by MediaSense makes me expect the tracks to be of the same length (track duration is the same, track start timestamp is the same) I have no way of knowing to which end of the shorter track I need to add silence, the end or beginning. Calls are in G711 mu-law if that makes any difference.  
 
My suspicion is that when MediaSense is sending the raw tracks using HTTP chunk transfer coding as described in the API docs it is not appropriately including an initial or terminal “SILENCE=x” chunk extension which would pad out the shorter track and make the tracks align when muxed, but I cannot prove that.    
 
I appreciate your feedback, thanks in advance!

Daniel,

How far off are the tracks from one another? The timings reported in the MediaSense session data are based on SIP messaging, not on actual RTP media flow. There is likely to be a small mismatch, but it shouldn't be very much.

I want to make sure that you are actually receiving some SILENCE= tags, right? Most normal HTTP 1.1 clients automatically combine chunks and in doing so, they lose these inter-chunk tags. You probably need to code your HTTP client at the TCP level in order to capture these tags. You've probably gotten that, but I just want to make sure.

I will check on the start and end silence periods and get back to you.

Thanks,
Jeff W.

From: Cisco Developer Community Forums [mailto:cdicuser@developer.cisco.com]
Sent: Wednesday, February 06, 2013 7:36 AM
To: cdicuser@developer.cisco.com
Subject: New Message from Daniel Hatch in Cisco MediaSense (MediaSense) - Technical Discussion: Raw tracks not in sync

Daniel Hatch has created a new message in the forum "Technical Discussion": -------------------------------------------------------------- I'm having a problem where one of the raw tracks provided by MediaSense is not the same length as the other one. Therefore when I mux the tracks together into a single call track people are talking over each other and it is difficult/impossible to follow the call. Because the metadata provided by MediaSense makes me expect the tracks to be of the same length (track duration is the same, track start timestamp is the same) I have no way of knowing to which end of the shorter track I need to add silence, the end or beginning. Calls are in G711 mu-law if that makes any difference.

My suspicion is that when MediaSense is sending the raw tracks using HTTP chunk transfer coding as described in the API docs it is not appropriately including an initial or terminal “SILENCE=x” chunk extension which would pad out the shorter track and make the tracks align when muxed, but I cannot prove that.

I appreciate your feedback, thanks in advance!
--
To respond to this post, please click the following link: http://developer.cisco.com/web/mediasense/forums/-/message_boards/view_message/11617085 or simply reply to this email.

Jeffrey Wolfeld:
How far off are the tracks from one another?

I most egregious example I have found was about 20 seconds. Usually they are more in the range of 3-6 seconds.
Jeffrey Wolfeld:
I want to make sure that you are actually receiving some SILENCE= tags, right? Most normal HTTP 1.1 clients automatically combine chunks and in doing so, they lose these inter-chunk tags. You probably need to code your HTTP client at the TCP level in order to capture these tags. You've probably gotten that, but I just want to make sure.
 
We're using our own code to parse the HTTP response and I've been over it/stepped thru it a few times making sure that it appropriately inserts silence and is reading the tags. I doubt the bug is there.
 
Thanks,
Dan

Dan,
Are you saying that the track durations are 3-6 and in some cases 20 seconds off from each other, or from the duration reported by the session data?
In either case, we will need some logs to investigate this.  I will contact you outside the forum.

Thanks,
Jeff W.

That's the difference between the tracks, I haven't actually compared them to session duration.