home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.dcom.isdn
- Path: sparky!uunet!cs.utexas.edu!csc.ti.com!tilde.csc.ti.com!trdc000.trdc.ti.com!picone
- From: picone@trdc001.trdc.ti.com (Joe Picone)
- Subject: end-to-end digital on ISDN -- the real application
- Message-ID: <PICONE.93Jan7174958@trdc001.trdc.ti.com>
- Lines: 69
- Sender: usenet@trdc.ti.com
- Nntp-Posting-Host: trdc001
- Organization: Tsukuba Research and Development Center
- Date: Thu, 7 Jan 1993 08:50:01 GMT
-
- Thanks for all the helpful responses.
-
- The answers to my previous messages were that *no one* can guarantee
- end-to-end digital connections.
-
- Perhaps I should explain our real application and let you ISDN wizards
- out there solve the problems...
-
- I am interested in collecting large speech databases that represent
- good samples of the variation of speakers, telephones, channels, and
- the ambient conditions found in the "normal telephone network"
- (whatever that means these days). Of course, we cannot sample all
- dimensions of the problem (collect data in every city, from every
- carrier , etc.).
-
- Two scenarios we have used in the past are the following (vendors
- names are used just for informational and historical purposes - let's
- leave marketing issues out of this):
-
- 1. Voice Across America
-
- Callers, randomly selected throughout the US, dial an 800 number
- supplied by AT&T. These callers call from any phone they want (home
- phone, office phones, analog, digital, etc.) any time of the day or
- night.
-
- Our computer system, in Dallas, is connected to the local switch via
- an analog telephone interface (a Gentner TC-100). It answers the
- phone, prompts the subjects for things to say, and digitizes the
- incoming calls with a high quality A/D.
-
- Note that the data collected under this scheme has been very
- valuable for speech recognition research - even with all its obvious
- deficiencies.
-
- 2. Switchboard
-
- The scenario is comparable to that above with major changes in the
- hardware. A T1 line is terminated in our lab. We connect an
- Intervoice RobotOperator to the T1. Users call from all over the
- country using a toll-free number. The numbers, I believe, have
- been registered with MCI (or maybe Bellcore) to provide end-to-end
- digital transmission (within the MCI network at least - again I
- don't know the details of what MCI actually promised).
-
-
- Case 1 suffers from the fact that at the receiving site there is a fixed
- analog line from the local office to the data collection machine.
- This does impose an additional and unnecessary A/D conversion.
-
- Case 2 suffers from the fact that the data collection system is a PC
- and T1s are unmanageable for us novices.
-
- So, now that Sun Sparcstation 10s have ISDN interfaces built in,
- we can replace Case 2 with an ISDN interface. Sounds too good to
- be true...
-
- The question was raised: how does the data collected in such a manner
- compare to the T1 connection in Case 2. My instinct is that doing this
- type of data collection with ISDN will be as good, certainly no worse,
- than the T1 case, and definitely better than Case 1.
-
- Others have raised concerns that there might be, on the average,
- more re-conversions performed.
-
- ISDN gurus please comment...
-
-
- -Joe
-