Discussion:
Speech overlapping with espeak-ng
Samuel Thibault
2017-01-15 14:57:48 UTC
Permalink
Hello,
If you have speakup running with espeakup just review up or down the
screen with the 7 and 9 keys. You can feel the speech doesn't cut off
immediately and in fact it sounds like you are mixing the speech from
the previous line in the current line. It gives a sound like a shoosh
at the start of each keystroke.
Ok, I think I see what you mean.

I believe it's actually an effect of our ears :)

I have noticed that espeak-ng is way more "reactive" than espeak, in
that when espeakup uses espeak-ng, key press echo happens way more
immediately than with espeak. This is so that when ongoing speech is
interrupted by newer speech, there is no delay between the interrupted
speech and the new speech, and our ears mix the two. To get something
intelligible we need some delay, which happened to be introduced by
espeak only by luck, because it was taking time to start the speech. I
have uploaded to

http://people.debian.org/~sthibault/tmp/

a version of espeakup which introduces a 150ms delay after the cancel.
This should be small enough to keep reactivity, but long enough to
clearly separate speech. You can tune this value with the additional -c
parameter.

I'll try to introduce that in brltty too.

Samuel
Samuel Thibault
2017-01-15 15:19:55 UTC
Permalink
Post by Samuel Thibault
I'll try to introduce that in brltty too.
I have uploaded brltty packages on

http://people.debian.org/~sthibault/tmp/

too. The delay can be changed with the cancel_pause espeak parameter,
e.g.:

/sbin/brltty -S cancel_pause=150 blabla...

or

speech-parameters es:cancel_pause=150

Could people give a try to these packages, and see which value of the
cancel pause they prefer?

Samuel
Hart Larry
2017-01-15 15:30:12 UTC
Permalink
While its a different type of package, I notice this overlap in 1 of the newer
Voxin 1.6 in Vinux. Also while arrowing, speech rate seems to drop by 1 step
until you re-read the current line.
Hart
Samuel Thibault
2017-01-23 12:21:03 UTC
Permalink
Hello,

Could people really please test these brltty and espeakup packages?

In short: if I don't get feedback very soon, I'll have to revert them to
espeak for Stretch, which would be very sad.

If there is something preventing from doing the tests, please tell, I
can't guess it.

Since the migration to espeak-ng poses strong problems, we *have* to
fix them before the release. If people don't actually test the fixes,
there's no way we can be sure that things are alright. I did test these
packages, but since I'm no speech user, I can't have a good idea of what
*sounds* right.

We have to decide very soon what to do for the next Debian release,
Stretch. The full freeze is 5th February, we can't introduce big
changes after that date, plus the 10-day migration period, it basically
means we have to decide by 25th January. If I don't get reports saying
that brltty 5.4-5 and espeakup 0.80-4 are now good, I'll revert to
espeak by 25th January.

To make testing yet easier, I have uploaded the packages to be tested
(brltty 5.4-5 and espeakup 0.80-4) into experimental. They are now
already available with:

deb http://incoming.debian.org/debian-buildd buildd-experimental main

Please test!

Samuel
I have uploaded to
http://people.debian.org/~sthibault/tmp/
a version of espeakup which introduces a 150ms delay after the cancel.
This should be small enough to keep reactivity, but long enough to
clearly separate speech. You can tune this value with the additional -c
parameter.
I have uploaded brltty packages on
http://people.debian.org/~sthibault/tmp/
too. The delay can be changed with the cancel_pause espeak parameter,
/sbin/brltty -S cancel_pause=150 blabla...
or
speech-parameters es:cancel_pause=150
Could people give a try to these packages, and see which value of the
cancel pause they prefer?
Samuel
Samuel Thibault
2017-01-23 12:31:24 UTC
Permalink
Post by Samuel Thibault
deb http://incoming.debian.org/debian-buildd buildd-experimental main
to be more precise: after having added this line to
/etc/apt/sources.list, run

apt-get update

apt-get install brltty=5.4-5

apt-get install espeakup=1:0.80-4

Samuel
Samuel Thibault
2017-01-23 13:07:48 UTC
Permalink
Post by Samuel Thibault
Post by Samuel Thibault
deb http://incoming.debian.org/debian-buildd buildd-experimental main
to be more precise: after having added this line to
/etc/apt/sources.list, run
apt-get update
apt-get install brltty=5.4-5
I meant brltty-espeak=5.4-5 of course, the brltty package does not
contain the espeak driver, the brltty-espeak package does.

Samuel
Tom Fowle
2017-01-24 04:31:27 UTC
Permalink
Am I correct to assume these versions do not effect use of speakup with
hardware synthesizers?
and would not apply to such setups using wheezie?
Thanks
Tom Fowle
Post by Samuel Thibault
Post by Samuel Thibault
Post by Samuel Thibault
deb http://incoming.debian.org/debian-buildd buildd-experimental main
to be more precise: after having added this line to
/etc/apt/sources.list, run
apt-get update
apt-get install brltty=5.4-5
I meant brltty-espeak=5.4-5 of course, the brltty package does not
contain the espeak driver, the brltty-espeak package does.
Samuel
_______________________________________________
Speakup mailing list
http://linux-speakup.org/cgi-bin/mailman/listinfo/speakup
Samuel Thibault
2017-01-24 08:21:15 UTC
Permalink
Post by Tom Fowle
Am I correct to assume these versions do not effect use of speakup with
hardware synthesizers?
Yes.
Post by Tom Fowle
and would not apply to such setups using wheezie?
It doesn't either.

Samuel
Øyvind Lode
2017-01-23 13:12:34 UTC
Permalink
Hi:

I've upgraded both brltty and espeakup (remotely via ssh since I'm at work now).
I'll help you out test later today.

What exactly should I test?

I vaguely remember seeing a thread about some issues but I must say
that I never noticed it myself.
Post by Samuel Thibault
Post by Samuel Thibault
deb http://incoming.debian.org/debian-buildd buildd-experimental main
to be more precise: after having added this line to
/etc/apt/sources.list, run
apt-get update
apt-get install brltty=5.4-5
apt-get install espeakup=1:0.80-4
Samuel
_______________________________________________
Speakup mailing list
http://linux-speakup.org/cgi-bin/mailman/listinfo/speakup
Øyvind Lode
2017-01-23 20:40:01 UTC
Permalink
Hi:

I've just tested by logging in on the console.
To me all works as expected.
That said I rarely login on the console since I use Debian as a server
and do all admin work via ssh which means I really don't know whether
what I get now on this server running sid is normal or not.
However, speech on the console is very useful when I really need it
like if network issues etc.
I typed commands without issues.
I read log files.
I opened nano and I type really fast and of course I couldn't hear
every character entered but I've never been able to do that when
typing really fast.
I edited the text and I also opened some real config files and did some editing.
Sorry for not finding/hearing the speech overlap.
Hi Øyvind
Post by Øyvind Lode
I've upgraded both brltty and espeakup (remotely via ssh since I'm at work now).
I'll help you out test later today.
Please also upgrade brltty-espeak.
Post by Øyvind Lode
What exactly should I test?
- Do you realize some kind of overlap when typing quickly?
- When navigating to a longer line, press Ctrl, do you recognize a delay
before the speech stops?
- You have to have a keyboard table configured for this, put
`keyboard-table laptop` into `/etc/brltty/brltty.conf` for this.
- When scrolling to a lot of text, say an inbox of Mutt, do you have the
feeling that speech overlaps?
- There is a parameter which introduces a certain cancelling pause, put that
`speech-parameters es:cancel_pause=150`
150 is the default value, please try different values (lower means less
pause), to figure out whether everything sounds as expected.
For me, a value of e.g. 10 works really fine when typing, but when scrolling
through many lines of text, it still sounds weird, as if it would overlap. As
Samuel already pointed out on IRC and here, it might be not an actual overlap,
but one the ear percepts.
At the moment, I feel as if a really small cancel pause would be the option I
could live best with.
Thanks
Sebastian
--
Web: http://www.crustulus.de (English|Deutsch) | Blog: http://www.crustulus.de/blog
FreeDict: Free multilingual dictionaries - http://www.freedict.org
Freies Latein-Deutsch-Wörterbuch: http://www.crustulus.de/freedict.de.html
Samuel Thibault
2017-01-24 02:36:43 UTC
Permalink
Hello,

Thanks to Sebastian's tests and reports (we ended up recording what was
produced with his audio setup), I could track and find what I believe
was the bug.

So, can people please test libespeak-ng1 version 1.49.0+dfsg-7 which
should be available within a few hours? You can use the unstable
versions of espeakup (1:0.80-5) and brltty (5.4-6), do not use the
experimental versions which introduce a delay which is actually
spurious.

I hope this time it'll be good enough for Stretch :)

The technical details are in
https://github.com/espeak-ng/espeak-ng/issues/172#issuecomment-274682363
Basically, espeak used to close/reopen the audio card, which espeak-ng
didn't, and some drivers or cards or something else doesn't really flush
data if that's not done.

I have to apologize to Kirk notably, it seems that depending on audio
setups there indeed was real overlapping, not only ears glitches :)

Samuel
Al Sten-Clanton
2017-01-24 02:40:01 UTC
Permalink
When do you need test results by? Is it tomorrow, the 26th? Thanks!
Post by Samuel Thibault
Hello,
Thanks to Sebastian's tests and reports (we ended up recording what was
produced with his audio setup), I could track and find what I believe
was the bug.
So, can people please test libespeak-ng1 version 1.49.0+dfsg-7 which
should be available within a few hours? You can use the unstable
versions of espeakup (1:0.80-5) and brltty (5.4-6), do not use the
experimental versions which introduce a delay which is actually
spurious.
I hope this time it'll be good enough for Stretch :)
The technical details are in
https://github.com/espeak-ng/espeak-ng/issues/172#issuecomment-274682363
Basically, espeak used to close/reopen the audio card, which espeak-ng
didn't, and some drivers or cards or something else doesn't really flush
data if that's not done.
I have to apologize to Kirk notably, it seems that depending on audio
setups there indeed was real overlapping, not only ears glitches :)
Samuel
_______________________________________________
Speakup mailing list
http://linux-speakup.org/cgi-bin/mailman/listinfo/speakup
Samuel Thibault
2017-01-24 02:46:55 UTC
Permalink
Post by Al Sten-Clanton
When do you need test results by? Is it tomorrow, the 26th? Thanks!
By 25th, please :)

Thanks,
Samuel
Al Sten-Clanton
2017-01-24 03:29:09 UTC
Permalink
I'll try to do it.

Al
Post by Samuel Thibault
Post by Al Sten-Clanton
When do you need test results by? Is it tomorrow, the 26th? Thanks!
By 25th, please :)
Thanks,
Samuel
_______________________________________________
Speakup mailing list
http://linux-speakup.org/cgi-bin/mailman/listinfo/speakup
Kirk Reiser
2017-01-24 18:06:16 UTC
Permalink
Hello Samuel and all: I have to say the fixes you put in over the past
day or so have improved espeak-ng responsiveness emensely. I've only
done preliminary tests so far but it seems to behaving itself very
well indeed. Almost instant flushing and clean sounding moving between
lines with the review keys. I didn't think you were going to pull it
off but you've done a very nice job Samuel. Thank you for working on
this so diligently.

Kirk
--
Well that's it then, colour me secure!

-----BEGIN PGP PUBLIC KEY BLOCK-----
Version: GnuPG v1
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=nrPH
-----END PGP PUBLIC KEY BLOCK-----
Samuel Thibault
2017-01-24 19:03:42 UTC
Permalink
I have to say the fixes you put in over the past day or so have
improved espeak-ng responsiveness emensely.
Cool :)

Samuel

Mark Peveto
2017-01-23 20:22:24 UTC
Permalink
Samual,
I don't have a lot in the way of technical knowledge, but I can try to help you test. What do you need? I don't use brltty, so if that's what you're
lookin for I might not be much help.


Mark Peveto
Registered Linux user number 600552
Everything happens after coffee!
Post by Samuel Thibault
Hello,
Could people really please test these brltty and espeakup packages?
In short: if I don't get feedback very soon, I'll have to revert them to
espeak for Stretch, which would be very sad.
If there is something preventing from doing the tests, please tell, I
can't guess it.
Since the migration to espeak-ng poses strong problems, we *have* to
fix them before the release. If people don't actually test the fixes,
there's no way we can be sure that things are alright. I did test these
packages, but since I'm no speech user, I can't have a good idea of what
*sounds* right.
We have to decide very soon what to do for the next Debian release,
Stretch. The full freeze is 5th February, we can't introduce big
changes after that date, plus the 10-day migration period, it basically
means we have to decide by 25th January. If I don't get reports saying
that brltty 5.4-5 and espeakup 0.80-4 are now good, I'll revert to
espeak by 25th January.
To make testing yet easier, I have uploaded the packages to be tested
(brltty 5.4-5 and espeakup 0.80-4) into experimental. They are now
deb http://incoming.debian.org/debian-buildd buildd-experimental main
Please test!
Samuel
I have uploaded to
http://people.debian.org/~sthibault/tmp/
a version of espeakup which introduces a 150ms delay after the cancel.
This should be small enough to keep reactivity, but long enough to
clearly separate speech. You can tune this value with the additional -c
parameter.
I have uploaded brltty packages on
http://people.debian.org/~sthibault/tmp/
too. The delay can be changed with the cancel_pause espeak parameter,
/sbin/brltty -S cancel_pause=150 blabla...
or
speech-parameters es:cancel_pause=150
Could people give a try to these packages, and see which value of the
cancel pause they prefer?
Samuel
_______________________________________________
Speakup mailing list
http://linux-speakup.org/cgi-bin/mailman/listinfo/speakup
Loading...