*start*
02172 00024 US 
Date:  8-Mar-83 16:10:33 PST
From: Ousterhout.pa
Subject: More on Transmission Protocols
To: VoiceProject^

Maybe there are other ways to solve the problems I mentioned in my
last message.  It turns out that there are 192 unused bytes in each
chirp that is stored on disk.  I could use 125 of those bytes to store a
silence map for the chirp, where each bit of the silence map
corresponds to 8 bytes (1 ms) of voice and indicates whether or not
that chunk of voice is silent.  Bluejay can then use the silence map
to be smart on playback and can handle variable-size packets.  The
silence map has other potential uses also:  Bluejay can keep better
track of the end of the recording (right now it only keeps track to the
nearest 1 sec. boundary), and perhaps the silence map might be of use
as a condensed representation of a recording (too bad that the real
granularity of silences will be the Lark packet size, rather than 8 bytes).

While I'm on the air, I propose that we eliminate the "packets sent" words
from Lark<->Bluejay packets.  Instead, I suggest that each packet contain
a "silence amount" field.  In packets from Lark to Bluejay, this word indicates
how many milliseconds of silence were not sent;  the silence occurred in time
just before the voice in the packet.  This value can be used by Bluejay to
figure out whether or not it lost some packets.  In packets from Bluejay to
Lark, the "silence amount" field indicates how many milliseconds of silence
are to FOLLOW the voice in the packet.  Once again, this information is
sufficient for Lark to figure out when it has missed something.  I assume that
in the event of long silences both Bluejay and Lark will occasionally send
packets with no data and only silence in order to let the other side know
where things stand.  I think that the scheme is powerful enough to permit
retry, and much simpler than the "packets sent" scheme.  In particular, it
avoids problems of mapping packet numbers into time intervals:  everything
can be done entirely in terms of the millisecond time-of-day.  I doubt that
a packet sequence number is even necessary in this scheme.

What do you think?*start*
01069 00024 US 
Date: 8 March 1983 4:47 pm PST (Tuesday)
From: ornstein.PA
Subject: Text to Speech Demo
To: VoiceProject^
Reply-To: ornstein

Eric Rawson sent me the following:

- - - - - - - 
Severo, are you aware of the "Type-N-Talk", a small (6" x 4" x 2") box which
also "synthesize[s] comprehensible speech algorithmically from arbitrary English
text"?  It costs about $300, accepts ascii strings on a standard RS-232 line, and
speaks over a loudspeaker.  I have one in my optics lab; I use it for speaking
verbal prompts at me when I'm running a computer-controlled optics experiment
(sometimes literally in the dark, when it is inconvenient to watch a CRT screen).
- - - - - - - 

Result is we have a date as follows (msg. from Rawson):

- - - - - - - 
3:30 PM Wed. 16th is OK.   Rm. No. of my office is 1550; lab is Rm. 1512.  Meet
me in my office, but if I'm not there I'm probably in that lab.  Phone No. is
unchanged (X4222).
- - - - - - - 

I hope that's OK with you (Larry and Dan). Assuming it is, we'll go right after
Dealer - next Wed.

Severo

*start*
01065 00024 US 
Date: 11 March 1983 10:35 am PST (Friday)
From: ornstein.PA
Subject: Etherphone $'s
To: Mitchell, Taylor, Lampson
cc: VoiceProject^, Overton
Reply-To: ornstein

We presently have 10 expensed Etherphones in the works and $60K capital in
our budget.  It seems that the price of an Etherphone has worked its way up to
about $1500 apiece - against our original guess of $1000.  (The reasons are just
what you expect - combination of too low initial guess, slight design
grandification, and a few expensive parts.)  Our inclination is to spend about
$50K of the $60K - reserving $10K for possible purchases - e.g. the
text-to-speech box.  $50K will buy a little over 30 more Etherphones giving us a
total of 40+.  That doesn't completely blanket the lab but is certainly plenty to
allow us credible trial useage.

ISL has indicated interest in a few.  I suggest we just place two or three of this
40+ in ISL and let them pay us back out of the next build when we are ready to
finish filling up CSL - and, presumably, ISL. 

Sound reasonable?

S.
*start*
00909 00024 USm
Date: 11 March 1983 10:57 am PST (Friday)
From: ornstein.PA
Subject: Etherphone 
To: Shoch 
cc: VoiceProject^, Taylor
Reply-To: ornstein

John,

You should know about the Etherphone - a reasonably powerful single board,
dual-8088 system.  It's good for a lot more than just voice and features: 

	64K bytes of shared RAM
	8K bytes of eprom for each of the processors
	8K bytes of Ram for the "slave" 8088
	Timer chip 
	DMA
	Priority Interrupts
	Watchdog timer
	Encryption

EXTERNALS:

	1.5 MB Ethernet (a Dicentra could gate this to either 3MB or 10MB net)
	Two RS-232 lines (SDLC capability)
	24 General Purpose I/O lines
	A buffered version of the shared 8088 bus 

It is a cheap (~$600) network-based I/O controller.  It's in PC form and we're
building 40 or so of them for starters.

Do you, or someone working for you, want more detail? (No documentation -
verbal only). 

Severo
*start*
00484 00024 US 
Date: 11 Mar 1983 13:24 PST
From: ornstein at PARC-MAXC
Subject: Increasingly official
To: Lia@SU-Score
cc: VoiceProject^
Reply-To: ornstein

Hi Lia,

This is just to let you know that the next hurdle has been passed and you are
specifically attached to the Voice Project for the summer.  Welcome!!  

An official offer letter should eventually come your way.  If you don't hear and
grow anxious, don't hesitate to call me or send a message.

Cheers,

Severo

*start*
00955 00024 US 
Date: 15 March 1983 12:01 pm PST (Tuesday)
From: Stewart.PA
Subject: Lark DIP switches
To: VoiceProject^
cc: Stewart

There are two 8 bit DIP switches on the printed circuit Etherphone Digital Board.

The switch at position K1 controls the Ethernet host address and two bits of Mode.  The switch at position F10 handles miscillaneous control signals and the third bit of Mode.

K1:

Switch Label		Function		Setting
-----------------------------------------------
8			ID0	(LSB)
7			ID1
6			ID2
5			ID3
4			ID4
3			ID5
2			Mode.0		Open
1			Mode.1		Closed

The Ethernet host address is the value of the ID field plus 100 octal.  ID0 is the least significant bit (the 1 bit).  ID5 is the 40 bit.


F1

Switch Label		Function		Setting
-----------------------------------------------
8			Mode.2		Open
7			ManNMI		Open
6			TimerInt		Closed
5			WDTOut'		Open
4			3Mb SLC		Open
3			1.5 Mb SLC		Closed
2			ClkDB4		Closed
1			ClkDB2		Open



*start*
00910 00024 US 
Date: 22 March 1983 9:39 am PST (Tuesday)
From: ornstein.PA
Subject: Msg. Requesting Alpha Users
To: Swinehart, Stewart
cc: Ornstein

I want to send roughly the following.  Any suggestions before it goes?

S.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

Subject: Etherphone Alpha Users?
To: CSL^

We are now soliciting volunteers to try out Etherphones*.  We need about ten
people.  We hope to put the first system into service on June 1.  It will provide
standard phone service with a few additional features (primitive voice message
service and work-station control).

Be the first on your block to lay encrypted Bitspeak on your colleagues.  Sign up
now.  

----------
*Please specify color desired. Available in pale blue, pale blue, or pale blue.  

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - 

*start*
00594 00024 US 
Date: 22 March 1983 5:45 pm PST (Tuesday)
From: ornstein.PA
Subject: Etherphone Alpha Users?
To: CSL^
Reply-To: ornstein

We are now soliciting volunteers to try out Etherphones*.  We need about ten
people and have six so far.  We hope to put the first system into service June 1. 
It will provide standard phone service plus primitive voice message service and
work-station control.

Be the first on your block to lay encrypted Bitspeak on your colleagues.  Sign up
now.  

----------
*Please specify color desired. Available in Ether blue, Pale blue, or Sky blue.  


*start*
01302 00024 US 
Date: 23 March 1983 10:03 am PST (Wednesday)
From: ornstein.PA
Subject: Re: Voice Project
In-reply-to: Your message of Tue, 22 Mar 83 20:53:12 PST
To: Lia Adams <Lia@SU-SCORE.ARPA>
cc: VoiceProject^
Reply-To: ornstein

Lia,

1.  If they DON'T offer you an internship, they'll have to deal with me - and
they know from past experience that that is not in their best interest.

2.  I don't know about the salary but that'll be in the offer letter.  I can probably
find out now if it matters.

3.  I'm sure the dates are negotiable.  What's your preference?  Mine would be to
have you start as early as you like and leave the far end open (although
perhaps you would want a "no earlier than foo" statement - so that you would
have a guaranteed minimum).

Let me know if you want me to do anything about any of this.  Else we'll wait
and see if they don't move at a suitable pace.

Glad to chat with you whenever you're ready.  Dan is redoing the Thrush
Server considerably and Larry has been cleaning up the Lark (Etherphone) code,
has the new voice-transmission protocols working, and is in the process of
building them into the Bluejay file server.  Our own server machine is due
shortly and we're still hoping to have the first set of real users in June. 

Cheers,

Severo  



*start*
01318 00024 US 
Date: 29 March 1983 12:42 pm PST (Tuesday)
From: ornstein.PA
Subject: Etherphone Alpha-Users
To: Swinehart, Stewart, Horning, Rovner, Mitchell, Cattell, Levin, Ornstein,
 McGregor, Kay, Taylor, Overton

Greetings:

First - thank you for your response.

Second - here is a summary of where we stand:  

1.	Swinehart 
2.	Swinehart (development))
3.	Stewart
4.	Stewart (development)
5.	Lia Adams  (development - Voice summer student)
--------------------
6.	Horning
7.	Rovner
8.	Mitchell
9.	Cattell
10.	Levin
--------------------
11.	Ornstein  - not a regular Cedar user
12.	McGregor - no 1.5 MB Ethernet cable in ISL
13	Kay - not a regular Cedar user
14. 	Taylor  - button problem and no Cedar capability
15. 	Overton  - no Cedar capability
--------------------

The last 5 people listed have the indicated special problems.  One can use an
Etherphone without Cedar, but it is awkward to use as anything more than a
regular phone.  There are special motivations for wanting one in Bob's office -
and perhaps mine.  I'm not sure exactly how many we will end up having
(working prototypes etc.) but the first ten are quite certain.  As the questions
clear up, I'll report further.

Once again - many thanks to all of you.  It should be fun and will become more
so as development proceeds. 

Severo
*start*
01104 00024 US 
Date: 6 April 1983 5:10 pm PST (Wednesday)
From: ornstein.PA
Subject: Etherphone Alpha-Users
To: Swinehart, Stewart, Horning, Rovner, Mitchell, Cattell, Levin, Ornstein,
 McGregor, Kay, Taylor, Overton

I have decided on the set of Alpha-Users.  They will be:

1.	Swinehart 
2.	Swinehart	(development)
3.	Stewart
4.	Stewart	(development)
5.	Lia Adams	(Voice summer student)
6.	Horning
7.	Rovner
8.	Mitchell
9.	Cattell
10.	Levin
11.	Ornstein

(We'll have 11 instead of 10 because Stewart will use the breadboard machine as
one of his two). 

To the rest of you - thanks for your willingness and hang in there.  We're
working on the next build.

Over coming weeks Mike Overton will ask your color preference, where you
want things located, etc. and will be gradually preparing the installations.  As
the new telephone instrument you will be getting will replace your present one,
we won't complete the installations until we are on the air and ready to go. 
Target is still 1 June.  The project members are pre-Alpha-Users and will be
using the system during May.

Cheers,

Severo

*start*
00294 00024 US 
Date: 11 April 1983 10:20 am PST (Monday)
From: ornstein.PA
Subject: Files archived from directory <VOICE>
To: VoiceProject^
Reply-To: ornstein

CSOFTWARE0383.DM;1 307 pages
LARKSLAVE0882.DM;1 32 pages
VOICEPARTITION0383.DM;1 291 pages
           3 files     630 pages

*start*
00759 00024 US 
Date: 11 Apr 83 21:37:38 PST (Monday)
From: Marzullo.pa
Subject: "Sentances We Hated to Come to the End Of"
To: Whimsy^
Reply-To: Marzullo.pa

From New Yorker February 14, 1983, p. 46:

"To the contrary, precisely because of the contextural (i.e. complexural) relatedness among selves, selves are immanently (i.e. reflexively) presenced to one another by way of their mutuality, their enabling/enabled relationships - i.e. "spirit" as "empowering" of self by the self is the very texture of the "self-relatedness" of the relation that relates itself to its own self and by so relating relates itself to the empowering other self, and, of course, conversely".

This excellent fogbank is from "The Context of Self" by Richard M. Zaner.

*start*
01777 00024 US 
Date: 14 April 1983 4:27 pm PST (Thursday)
From: ornstein.PA
Subject: Voice Paper?
To: Taylor
cc: Swinehart, Stewart, ornstein

We have been invited to give a paper at the fall (Nov-Dec) meeting of
GlobeCom83* in San Diego. The session will be on special local-area-network
services.  They are interested in particular in voice comminication over the
Ethernet.  Prof. Thomas E. Stern**, Dept. of EE, Columbia Univ. New York, NY
10027 was pointed at us by John Shoch.  I promised I'd write and tell him yes or
no next week. (If yes, he wants title and author's names). 

This looks like a good conference to present our project at and we would have
time to write a good paper.  (We can by-pass the normal review process and thus
would probably have until August sometime to produce the final copy.)   But I
have some concern that while the basic system will be operating by then, we
will not yet have much user or performance information, nor will we have
explored many of the various user capabilities we hope to provide.  It's not lack
of material that concerns me, we have enormous amount of good stuff to report. 
It's just that it's mostly underlying architecture and facilities, rather than neat
user results.

Larry and Dan are inclined to go for it and that carries a lot of weight with me 
so I guess I'm inclined to go ahead.  Do you have any opinion or forsee any
snags - for example, is it likely we will get into any security problem?

-----------------------
*There have been two main telecommunications meetings: ICC (International
Communications Conference), which meets in the spring - and NTC (National
Telecommunications Conference which meets in the fall.  GlobeCom is NTC
renamed.

**(212) 280-3123 (office)
   (212) 249-0479 (home)
*start*
00527 00024 US 
Date: 14-Apr-83 17:18:37 PST
From: Swinehart.pa
Subject: Authentication Server
To: Birrell
cc: Stewart, Swinehart

Reminder that it would be nice to have a real live authentic authentication server (or two) running in the internetwork.  Even nicer if it could be an RPC-based one.  Would it be possible to produce a Cedar-based RServer whose major responsibility would be to duplicate registries for the purpose of supporting such a facility?  Could run it on any of our burgeoning Cedar-based servers.
*start*
00855 00024 US 
Date: 14-Apr-83 17:22:35 PST
From: Swinehart.pa
Subject: Transmitting the past -- a proposal
To: Stewart
cc: VoiceProject^

A reminder of my brainstorm that might improve the perceived quality of voice transmissions at the end of a silence interval:  since the recipient is going to wait a half packet time before playing the first non-silent packet, it might as well play the last half of the previous, allegedly silent, packet.  The sender would have to send these two packets back to back, or even send a 50% longer packet the first time, with a bit of the past in it.

If all senders did this, the half-packet delay code could be removed from the receive routines, which would just start playing as soon as they received anything.

You mentioned you might need a special packet type to deal with these things.  Sounds OK.

Dan
*start*
02233 00024 US 
Date: Wed, 27 Apr 83 10:21 PDT
From: Stewart.PA
Subject: Another text to speech unit
To: VoiceInterest^.pa
cc: Stewart

Forwarded from HUMAN-NETS

From: "REX::MINOW c/o " <DEC-HNT at DEC-MARLBORO>
Date: 13-APR-1983 11:27
Subject: For a good time, call (617) 493-7625

For the last year, I have been working on DECtalk, a high quality
text-to-speech synthesizer.  An article on this work will appear in
next week's Electronics (should be out the week of April 18th) and I
thought Human Nets, Telecom and USEnet people might enjoy a brief
summary and an early chance to hear the demo.

DECtalk uses a 68000 with 256K-bytes ROM to convert unrestricted
English text to synthesizer parameters.  These are transmitted to a
TMS32010 digital signal processor to generate the analog waveform.
The 68000 uses a large lexicon and a set of about 400
letter-to-sound conversion rules.  The lexicon (occupying about 1/2
of the ROM space) guarantees correct pronunciation for a large
subset of English and cuts down processing time for common words.
DECtalk also contains heuristics to process abbreviations, numbers,
and acronyms.

To communicate with the outside world, DECtalk contains two
asynchronous terminal lines (one with modem control) and a built-in
telephone line interface with DTMF (Touch-Tone) decoder.

The processing requirements are interesting:  English text, entered
at about 30 bytes per second, is converted by the 68000 to
synthesizer parameter blocks (18 16-bit words).  The TMS DSP reads a
new parameter block every 6.5 msec. and generates 10,000 12-bit
samples per second.  All processing is digital -- the only analog
component on the board is the DAC anti-aliasing filter.

DECtalk software was written in C -- the board has a home-brew
Unix-flavored real-time operating system.  This allowed us to debug
text-to-speech modules a timesharing system.  (The non-real-time
components run on VMS, Unix, RSTS, RT11, and TOPS-20.)

If you would rather listen to DECtalk than read about it, feel free
to call (617) 493-7625 (preferably from a Touch-Tone phone).

Your comments and suggestions are most welcome.

Regards.


Martin Minow
decvax!minow            (USENET)
decvax!minow @ Berkeley (ARPA)



*start*
00757 00024 US 
Date: 2 May 1983 12:02 pm PDT (Monday)
From: ornstein.PA
Subject: Voice Paper
To: Taylor
cc: Swinehart, Stewart, ornstein, Lampson

I spoke to Butler re. possible proprietary problems with our agreeing to give a
Voice Paper at the GlobeCom83 conference in San Diego in the fall.  He saw no
problem with it.  Our proposed title would be:

		"Adding Voice to an Office Computer Network"

The paper would describe what we have done so far - the broad architecture,
motivations, highlights of the specific component designs (Ertherphone, Server,
Voice-File-Server, etc.)

1.  OK?

2.  Should I cheack with anyone else (Gloria?) who will eventually be in the
clearance loop before committing?

My committment is already overdue.

Severo

*start*
00587 00024 US 
Date: 2 May 1983 4:17 pm PDT (Monday)
From: ornstein.PA
Subject: Voice Paper Confirmed
To: Stewart, Swinehart
cc: Taylor, ornstein

Taylor and Spencer have both now given approval to go ahead, so I called Stern
and agreed we would give a paper entitled:

		"Adding Voice to an Office Computer Network"

All three of us are authors and the paper is due to them in finished form by
August 15.  We should give him a draft somewhat earlier.  (Camera ready copy
goes to the printer August 31).   We should start working on this in the latter
half of June I'd say.

S.
*start*
00733 00024 US 
Date: 11-May-83 16:40:27 PDT
From: Swinehart.pa
Subject: B..ken up ...ther....ne c...sati...s
To: L...ry ...tew... <Stewart.PA>
cc: Ornstein, Swinehart

In an otherwise wonderful front door conversation this morning (SSilThresh_240H), Severo and I encountered periods of near-complete loss of signal -- probably 70% or 80% of the packets were lost, or the equivalent.  It seemed to happen after one or both of us had been talking steadily for some time (probably Severo), and it seemed to alleviate when we both stopped for an instance.  Once it started, the breakup occurred in both directions, although I have the impression it didn't start at the same time in both directions.

So here's the thing ...

Dan*start*
13399 00024 US 
Date: Fri, 13 May 83 16:44 PDT
From: Stewart.PA
Subject: Back door levels
To: VoiceProject^.pa
cc: Stewart

Here is the strategy for testing the back door levels.

General comments:

Inside the analog board, all sources should produce "the same" level and all sinks should accept this level.  This is easy for codecs and line.  However, because of the loss in the local loop, telephone sets are designed to transmit a relatively louder signal and receive a relatively weaker one.

The amplifiers which handle the telephone set and the telephone line will have to compensate for this effect.

The TeleSetSource amplifier will introduce loss to bring down the teleset transmit level to the standard level.  The TeleSetSink amplifier will introduce loss to reduce the standard level to that expected by the telephone set.

The TeleWallSource amplifier will introduce gain to boost the weak signal from the central office to the standard level.  The TeleWallSink amplifier will introduce gain to bring the standard level up so that it will compensate for the loss going to the central office.

This design strategy has a deleterious effect on hybrid performance.  Suppose the round trip loss to the CO is 10 dB.  Then the transmit and receive arms of the four wire side of our hybrid must have 5 dB gain each.  If the native hybrid had 20 dB return loss, then the final system will have only a 10 dB return loss.

What is the standard level?

According to CCITT, a test tone at 0 dBm analog should be 3.17 dB below the digital clipping level.  Touch tones are (I think) transmitted at 0 dBM and voice levels have an average power of about -10 to -13 dBm.

Lets work with voltage rather than power, so that we can ignore impedance levels for the moment.  0 dBm is 1 milliwatt in 600 ohms, corresponding to  (P = E^2/R)  0.775 v RMS or 0.775 * 1.41 * 2 = 2.19 v peak-to-peak.

I've lost the message describing the original analog board level design, so lets reconstruct it from the schematics.  Digital clipping level at the output of the 2912 (pin 4) is +/- 3.2 v or 9.3 dBm.  The resistive divider (R7, R8) connecting the codec to the crossbar has a ratio of .333, putting the peak voltage at the crossbar at 2.13 v p-p.  Thus the codec clipping level is -0.24 dB and the voltage at the crossbar corresponding to 0 dBm is 3.17 dB lower or -3.41 dBm.  The voltages involved at the crossbar are .523 v RMS or 1.48 v p-p.

The transmit section of the codec (CodecSink) has an intrinsic 3 dB gain, plus gain setting resistors, the gain is (1 + R5/R4) + 3 dB.  R5 = R4, so the overall gain is 2.82 (1/0.354).

The line-out amplifiers have a gain of .687, (1/1/455) so that 0 dB at the crossbar produces about a 1 v p-p signal at line out.  Line in has a nearly inverse gain of 1.5, to restore the standard level.

TeleSetSource has a gain of 0.386 (or -8.26 dB).  After accounting for the -3.41 dB reference level, the extraordinary gain of TeleSetSource is -4.85 dB.

TeleSetSink has a gain of 1.33 (2.5 dB), but the Telephone set receiver is fed through a 910 ohm resistor, which forms a voltage divider together with the receiver element.  The division ratio of this voltage divider must be measured.

Aside on two wire local loops.

Consider a phone system in which only the local loop is two wire and that the connection is four wire inside the subscriber equipment and inside the central office, with a hybrid at each end.  Suppose further that the local loop has 5 dB loss in each direction.  The phone company can choose to have equal transmit and receive levels on its end of the local loop, thus getting the full return loss benefit of the CO end hybrid.  The subsciber must add gain to compensate for loop loss and consequently must put up with poorer hybrid performance.

How to get the resistors right.

Measure round trip loss for a local call.  Use differential scope to measure the voltage on tip and ring at the sending and receiving ends.  Measure the loss at several frequencies.  Half the loss in dB (square root of voltage ratio) can be assigned to the loss in each direction.  Call the one-way loss X dB.

Check: Call the tone generator number and measure the differential voltage at this end.  The value should be -X dBm.

Measure the (additional) loss introduced by the hybrid by measuring the single ended signal at the transmit port and the two wire differential signal at the phone line.  The rule is that

A Triplett 630 VOM was placed in series with tip and ring on 4478 to measure the DC loop currents drawn by various units.

Bell set from 4477   36 mA
Bell set from 4478 32 mA
ITT set (reverted) 32 mA
ITT set (reverted) with ringer shunted across loop   36 mA
TeleWall ckt   33 mA
TeleWall ckt with ringer shunted across loop   29 mA

I measured the RMS dial tone voltage across tip and ring.
p-p on scope .45
RMS (fluke) .123  (blue phone off hook)
RMS (fluke) .133  (black phone off hook)

This is -16 dBm.  For a single tone it would be -19 dBm.  According to notes on the Network, Dial tone is -13 dBm per frequency, giving a CO to here loss of 6 dB.

I made frequency dependent measurements at various points, with a phone call open between 4477 (a black phone) and 4478 (A Lark with it's TeleSet unplugged.)
p-p 4478 is peak-to-peak voltage measured between tip and ring at the 4478 (Lark) end
rms 4478 is RMS voltage measured between tip and ring at the 4478 (Lark) end
rms 4477 is RMS voltage measured between tip and ring at the 4477 end
rms twsrc is RMS voltage measured at TeleWallSrc inside the analog board
rms line1 is RMS voltage measured at Line1Src inside the analog board
dB loss is the power ratio in dB between the tip-ring differential voltages at the two ends of the phone connection (rms 4478 and rms 4477).
dB twTXgain is the power ratio in dB between rms line1 and rms 4478
dB hyb is the power ratio in dB between rms line1 and rms twsrc (hybrid return loss)

The dB loss entries represent the frequency specific end-to-end loss of the local telephone call.  Half this values should be the one way loss to or from the CO.  The figures are consistant with the measurements of dial tone level above.

  f     p-p   rms   rms   rms     rms   dB   dB           dB
        4478  4478  4477   twsrc  line1  loss  twTXgain  hyb
3000   1.48   .568   .092   .198   .374   15.8   3.6          -5.5
2000   1.41   .544   .126   .147   .374   12.7   3.3          -8.1
1500   1.32   .512   .134   .105   .374   11.6   2.7          -11.0
1000   1.36   .529   .142   .147   .374   11.4   3.0          -8.1
 800   1.44   .555   .142   .176   .373   11.8   3.5          -6.5
 600   1.52   .583   .139   .200   .372   12.5   3.9          -5.4
 400   1.56   .599   .132   .209   .371   13.1   4.2          -5.0
 300   1.55   .590   .116   .199   .370   14.1   4.1          -5.4
 
 I made measurements of rms voltage across tip and ring of a call placed between 4477 and 4478.
 The instrument at the 4477 end was a standard black phone.  The instrument at the 4478 end was a standard black phone.  In addition, there were on-hook and not-reverted Larks hung on both 4477 and 4478 (The ring detect circuits might present a small load.)
 
 The rows of the tables are DTMF buttons held down as signal sources.  Thus 4478  7,8 means that the 7 and 8 buttons were held down on the phone on line 4478.  The frequencies for button 7 are  852 and 1209 Hz.  The 7 and 8 buttons together generate only 852 Hz.
 
             rms      rms   dB
            4477     4478   gain
4477 7     .976      .272   -11.1
4477 7,8   .753      .201   -11.5
 
4478  7     .287      1.068   -11.4
4478  7,8   .202      0.774   -11.7

 The same configuration, except using the reverted Blue phone at the 4478 end with the black phone unplugged.
            rms       rms   dB
            4477     4478   gain
blue  7             .812
blue  7,8           .498
4477 7     .976     .274   -11.0
4477 7,8   .753     .199   -11.6

The blue phone has about the same audio load as the black phone
The blue phone generates lower level DTMF (by 1.8 dB)

The same configuration, except using the Lark hybrid as a receiver with the black phone unplugged and blue phone un-reverted.

rms xbar is the RMS voltage appearing at TeleWallSrc
dB twRXgain is the power ration between rms xbar and rms 4478

             rms      rms   rms   dB    dB    
            4477     4478   xbar  loss   twRXgain
4477 7     .976     .192    .252   14.1   2.4
4477 7,8   .753     .140    .184   14.6   2.4

I placed a call to 9-494-0020, which is a 1000 Hz test tone.  Probably the tone is at 0 dBm, but I am not sure.  RMS voltage measurements were made accross tip and ring with a load provided by the blue, black, and Lark hybrid for respective columns.  

meas.           blue  blk   EP    xbar
test tone rms    .48   .47   .315   .413    rms voltages at various points
test tone dBm  -4.2  -4.3  -7.8   -5.5     dBm equivalents

The ratio of EP and xbar = 2.4 dB, as above.
If the tone is 0 dBm at the CO, then the local loop loss is  about 4.2 dB.  I discount this, givent he good evidnece for loss in the 5.5 to 6 dB area above.


I placed a call from the Lark at 4478 to 4477 and measured the rms voltage accross tip and ring at 4478 generated by dtmf buttons on the blue phone, both reverted and via the electonics.

Measured at tip and ring
blue 7,8 via elec phone   0.225
blue 7,8 via revert      0.493

These values are 6.8 dB different.
After applying the fixes detailed below, the value was .464 via electronics)

I then measured the voltage accross the blue phone receiver element in reverted and electronic configurations. (The tones were generated by the DTMF pad 7 button at 4477).

electronic phone  .02
reverted           .079

These values are 9.6 dB different.
After applying the fixes detailed below, the value was .067 via electronics)

I placed a "front door" call from the blue phone to itself and measured the voltages across the receiver element generated by the DTMF pad.  (Column labelled rms elec.)
I also placed a reverted back door call to 4477 and measured the voltages generated by the 4477 DTMF pad (column labelled rms reverted).  These two columns use different DTMF pads, and the blue pad is weaker than most black pads (see above).

          rms   rms
          elec  reverted
7+8     .011   .047   (12.6 dB different)   (probably wrong??)
7        .023   .069   (9.5 dB different)    checks with above test

In order to check whether the electonic version was being insufficiently powered by low talk battery, I measured the voltage between the blue phones tip and ring in both reverted and non-reverted configurations for DTMF buttons on the blue phone.

          rms   rms
          elec  reverted
7+8     .523   .502
7        .882   .821

See above that DTMF levels are about tthe same if reverted or not, so dc drive currect is sufficient to supply telephone set

Notes:

1.   The Lark presents a smaller impedance to the telephone line than a standard desk set.  This means it gets less signal from the line.  I have not measured the impedances at audio frequencies but the Lark should be 600 ohms (that is the transformer rating) and the standard phones may be 900 ohms.

Suppose the phones are 900 ohms.  We can use the measured voltages at the two ends to estimate the loop resistance as though it were a single resistance.

The voltage at the sending end is 0.976 and the voltage at the receiving end is 0.272.  The difference, .704, is developed accross the loop resistance.  .704/Rl = .272/900, so Rl = 2300 ohms.  We can apply half the value for the resistance to the CO.

These numbers are consistant.  A .976 v source will develop .275 volts across 900 ohms at the other end of a 2300 ohm loop, while the same source will develop only .202 volts across a 600 ohm load.  Thus we have a received signal loss of 2.7 dB due to the impedance difference.

The change should not affect the transmit side much.  The lower drive impedance of the Lark should actually improve the transmit levels by 0.7 dB.

Conclusion, TeleWall source gain should be increased by 3 dB.

2.   The front door levels are lower than those of a back door call by about 9.5 dB.

Connecting TeleSet to TeleSet via the crossbar gives 2.0 dB more signal than connecting them via the codecs.

What accounts for the other 7.5 dB difference?  Is it the TeleSet source or sink or some combination?

The difference between electronic phone receive level and reverted receive level is 9.5 dB.  If 3 dB is due to the Telewall receive gain, then the other 6 is due to low TeleSetSink gain.  This computation would have the TeleSetSource gain low by 1.5 dB.

The difference between electronic phone transmit level and reverted transmit level is 6.8 dB.  If 2 dB is due to the Codec chain, then the remaining 4.8 must be split between TeleSetSource and TeleWallSink.  If TeleSetSource gain is low by 1.5 dB, then the remaining 3 dB would be due to TeleWallSink.

Proposal:

Boost TeleWallSource by 3 dB
	Change R70 from 27K to 47K
Boost TeleWallSink by 3 dB
	Change R62 from 11K to 18K
Boost TeleSetSource by 1.5 dB
	Change R76 from 390 ohms to 510 ohms
Boost TeleSetSink by 6 dB
	Change R82 from 24K to 47K

Boost CodecSink gains by 1.5 dB
	Change R5 from 22K to 30K
	Change R14 from 22K to 30K

I modified two analog boards and tested one of them, giving the (after fixes) numbers above.
*start*
01587 00024 US 
Date: 18 May 1983 4:46 pm EDT (Wednesday)
From: McIrvine.STHQ
Subject: MIT Program on Speech Input-Output 
To: VoiceInterest^.PA
cc: McIrvine.sthq
Reply-To: McIrvine

On May 17, 1983 I attended the MIT Industrial Liaison Program on "Issues
Relating to Speech Input-Output for Computers."  The program involved Dennis
Klatt, Victor Zue, and others.  It included presentations on Klattalk, DECtalk, and
MITalk, as well as a hatchet-job on the IBM Yorktown approach to Speech
Recognition.

I am a "hit-and-run" observer of speech recognition research.  (I first became
interested during a visit to the SRI project on speech recognition in about 1965. 
George M. White and Raj Reddy updated my knowledge circa 1973.  I have not
been in touch with the field since.)  With the resulting disclaimers regarding my
qualifications to comment on the current status in the field, here are my overall
reactions to the MIT program:

(1) Speech Recognition is just as difficult as I recall it to be.  It continues to be
dominated by parochial and heuristic approaches, just as it was 20 years ago. 

(2) Speech Synthesis is MORE difficult than I recall.  Improvement continues
through intelligent use of linguistic context to assist in determining the subtleties
of spacing, inflexion, and pitch.  However, the "state-of-the-art" is not up to my
aesthetic standards.  While certainly intelligible, DECtalk sounds disturbingly
like a Swede talking pretty good English. 

If anyone has any specific questions regarding the MIT Program, let me know
and I will try to answer.

Ted