Topics

changes in speech in NVDA 2019.3


Giles Turnbull
 

Hi all,

does anyone know whether there have been changes to the speech processing part of NVDA in the 2019.3 update? I didn't spot anything in the change log, but I experience a couple of things that are different from the 2019.2.1 version.

The first one is very minor ... when I tag an email in gmail so that it goes into an existing  folder, I press c to go to the combo box listing all folders, then I press a letter corresponding to the first letter of the folder name - such as f for my Facebook notifications. Once focus is on Facebook I tab over and activate the Go button, which applied the label to the email. Since 2019.3 the letter I pressed (in this case f) gets spoken when I tab over to the Go button; this did not happen before 2019.3.If I need to press the letter more than once to continue past Facebook to Farming and then to Fiction, NVDA now announces "fff" which is mildly annoying!

More of a problem is when I mistype words ... I regularly mistype because as becuase and words ending in tion as toin ... I suspect when I've done this and immediately select the word and retype it without presing space to move onto typing the next word. As a result the first mistyping doesn't get spoken straight away ... it gets spoken as I press space or punctuation ... and it merges the mistyped word in with the correct version, and they are seemlessly connected like they are one single word, becuasebecause ... the same happened in this post when I mistyped punctuation as puncutation and got "puncutationpunctuation" as a result.

The only irritation with this is that it makes it difficult to tell whether the subsequent attempts at spelling the word correctly has removed the typo or whether it is still mistyped.

I'm curious whether this is an intentional change and, if so, what the thinking behind it is.

Giles


Quentin Christensen
 

We did rewrite a lot of the speech code, although for now this shouldn't be an overly different experience, it was more to make future changes and maintenance easier.  Evidently you've found something that does work slightly differently.

Not sure about the first one, I'll have to learn GMails keyboard shortcuts more - but re editing words, I've written that up https://github.com/nvaccess/nvda/issues/10808

Regards

Quentin.

On Fri, Feb 21, 2020 at 2:40 AM Giles Turnbull <giles.turnbull@...> wrote:
Hi all,

does anyone know whether there have been changes to the speech processing part of NVDA in the 2019.3 update? I didn't spot anything in the change log, but I experience a couple of things that are different from the 2019.2.1 version.

The first one is very minor ... when I tag an email in gmail so that it goes into an existing  folder, I press c to go to the combo box listing all folders, then I press a letter corresponding to the first letter of the folder name - such as f for my Facebook notifications. Once focus is on Facebook I tab over and activate the Go button, which applied the label to the email. Since 2019.3 the letter I pressed (in this case f) gets spoken when I tab over to the Go button; this did not happen before 2019.3.If I need to press the letter more than once to continue past Facebook to Farming and then to Fiction, NVDA now announces "fff" which is mildly annoying!

More of a problem is when I mistype words ... I regularly mistype because as becuase and words ending in tion as toin ... I suspect when I've done this and immediately select the word and retype it without presing space to move onto typing the next word. As a result the first mistyping doesn't get spoken straight away ... it gets spoken as I press space or punctuation ... and it merges the mistyped word in with the correct version, and they are seemlessly connected like they are one single word, becuasebecause ... the same happened in this post when I mistyped punctuation as puncutation and got "puncutationpunctuation" as a result.

The only irritation with this is that it makes it difficult to tell whether the subsequent attempts at spelling the word correctly has removed the typo or whether it is still mistyped.

I'm curious whether this is an intentional change and, if so, what the thinking behind it is.

Giles



--
Quentin Christensen
Training and Support Manager



Giles Turnbull
 

thanks, Quentin :)


 

I have noticed with especially ends of words in espeak at any rate not sure with the fest but they seem to pop at the end of words and sound more abbrupt especially at the end of sentences.

It doesn't always happen but enough times a user would notice.

It doesn't mean the user can't use the software but you notice it when reading.



On 21/02/2020 5:02 pm, Quentin Christensen wrote:
We did rewrite a lot of the speech code, although for now this shouldn't be an overly different experience, it was more to make future changes and maintenance easier.  Evidently you've found something that does work slightly differently.

Not sure about the first one, I'll have to learn GMails keyboard shortcuts more - but re editing words, I've written that up https://github.com/nvaccess/nvda/issues/10808

Regards

Quentin.

On Fri, Feb 21, 2020 at 2:40 AM Giles Turnbull <giles.turnbull@...> wrote:
Hi all,

does anyone know whether there have been changes to the speech processing part of NVDA in the 2019.3 update? I didn't spot anything in the change log, but I experience a couple of things that are different from the 2019.2.1 version.

The first one is very minor ... when I tag an email in gmail so that it goes into an existing  folder, I press c to go to the combo box listing all folders, then I press a letter corresponding to the first letter of the folder name - such as f for my Facebook notifications. Once focus is on Facebook I tab over and activate the Go button, which applied the label to the email. Since 2019.3 the letter I pressed (in this case f) gets spoken when I tab over to the Go button; this did not happen before 2019.3.If I need to press the letter more than once to continue past Facebook to Farming and then to Fiction, NVDA now announces "fff" which is mildly annoying!

More of a problem is when I mistype words ... I regularly mistype because as becuase and words ending in tion as toin ... I suspect when I've done this and immediately select the word and retype it without presing space to move onto typing the next word. As a result the first mistyping doesn't get spoken straight away ... it gets spoken as I press space or punctuation ... and it merges the mistyped word in with the correct version, and they are seemlessly connected like they are one single word, becuasebecause ... the same happened in this post when I mistyped punctuation as puncutation and got "puncutationpunctuation" as a result.

The only irritation with this is that it makes it difficult to tell whether the subsequent attempts at spelling the word correctly has removed the typo or whether it is still mistyped.

I'm curious whether this is an intentional change and, if so, what the thinking behind it is.

Giles


--
Quentin Christensen
Training and Support Manager



Quentin Christensen
 

Shaun,

I haven't noticed that with eSpeak, I wonder if it is related to the sound card rather than NVDA - eg when an audio stream finishes (which would often be at the end of a sentence).  What sound card do you have, and what other relevant sound settings, eg audio ducking?

If you can find a way to reliably reproduce it and record it, that would be worth listening to.  I'm guessing there is nothing in the NVDA log to indicate an issue?  One last question - does it sound different if you are using speakers vs headphones?

Regards

Quentin.

On Wed, Feb 26, 2020 at 6:38 AM Shaun Everiss <sm.everiss@...> wrote:

I have noticed with especially ends of words in espeak at any rate not sure with the fest but they seem to pop at the end of words and sound more abbrupt especially at the end of sentences.

It doesn't always happen but enough times a user would notice.

It doesn't mean the user can't use the software but you notice it when reading.



On 21/02/2020 5:02 pm, Quentin Christensen wrote:
We did rewrite a lot of the speech code, although for now this shouldn't be an overly different experience, it was more to make future changes and maintenance easier.  Evidently you've found something that does work slightly differently.

Not sure about the first one, I'll have to learn GMails keyboard shortcuts more - but re editing words, I've written that up https://github.com/nvaccess/nvda/issues/10808

Regards

Quentin.

On Fri, Feb 21, 2020 at 2:40 AM Giles Turnbull <giles.turnbull@...> wrote:
Hi all,

does anyone know whether there have been changes to the speech processing part of NVDA in the 2019.3 update? I didn't spot anything in the change log, but I experience a couple of things that are different from the 2019.2.1 version.

The first one is very minor ... when I tag an email in gmail so that it goes into an existing  folder, I press c to go to the combo box listing all folders, then I press a letter corresponding to the first letter of the folder name - such as f for my Facebook notifications. Once focus is on Facebook I tab over and activate the Go button, which applied the label to the email. Since 2019.3 the letter I pressed (in this case f) gets spoken when I tab over to the Go button; this did not happen before 2019.3.If I need to press the letter more than once to continue past Facebook to Farming and then to Fiction, NVDA now announces "fff" which is mildly annoying!

More of a problem is when I mistype words ... I regularly mistype because as becuase and words ending in tion as toin ... I suspect when I've done this and immediately select the word and retype it without presing space to move onto typing the next word. As a result the first mistyping doesn't get spoken straight away ... it gets spoken as I press space or punctuation ... and it merges the mistyped word in with the correct version, and they are seemlessly connected like they are one single word, becuasebecause ... the same happened in this post when I mistyped punctuation as puncutation and got "puncutationpunctuation" as a result.

The only irritation with this is that it makes it difficult to tell whether the subsequent attempts at spelling the word correctly has removed the typo or whether it is still mistyped.

I'm curious whether this is an intentional change and, if so, what the thinking behind it is.

Giles


--
Quentin Christensen
Training and Support Manager




--
Quentin Christensen
Training and Support Manager



David Csercsics
 

I notice a similar issue with espeak where the beginning and end of utterances will sometimes clip with both of my USB sound cards. One of them is a AudioQuest DragonFly Red which is sometimes hooked to headphones and sometimes studio monitors. When I first start NVDA the first part of the startup tones gets clipped and the first syllable of the first utterance from espeak is clipped off as well. This happens less frequently with the Logitech h800 headphones I have, but even on those it still occurs occasionally.


Quentin Christensen
 

I wonder does it happen with other sounds as well, eg the start of a music track, the first ring of Skype, etc?

My guess is it's likely the driver - allowing the device to try to play sounds before it has properly initialised etc, and I doubt there is too much NVDA itself can do.

On Wed, Feb 26, 2020 at 11:22 AM David Csercsics <bleeblat@...> wrote:
I notice a similar issue with espeak where the beginning and end of
utterances will sometimes clip with both of my USB sound cards. One of
them is a AudioQuest DragonFly Red which is sometimes hooked to
headphones and sometimes studio monitors. When I first start NVDA the
first part of the startup tones gets clipped and the first syllable of
the first utterance from espeak is clipped off as well. This happens
less frequently with the Logitech h800 headphones I have, but even on
those it still occurs occasionally.







--
Quentin Christensen
Training and Support Manager



David Csercsics
 

No, I haven't seen it with other sounds unless the sound card was previously asleep for a long period. It's probably a driver bug somewhere. It's not too severe, just a bit strange. The sound card that glitches most frequently with NVDA only supports 24-bit audio, so maybe that has something to do with it.


 

No ducking or enhancements or anything like that.

Saying that there area few unique enhancements in the driver I can't turn off unless I kill it but if I do that the sound comes out as overdriven junk and I suspect some of this is to protect the speakers from the sheer power of the amp and stuff in the workstation.

Its not that bad, and well could be anything, no log errors or anything.



On 26/02/2020 12:54 pm, Quentin Christensen wrote:
Shaun,

I haven't noticed that with eSpeak, I wonder if it is related to the sound card rather than NVDA - eg when an audio stream finishes (which would often be at the end of a sentence).  What sound card do you have, and what other relevant sound settings, eg audio ducking?

If you can find a way to reliably reproduce it and record it, that would be worth listening to.  I'm guessing there is nothing in the NVDA log to indicate an issue?  One last question - does it sound different if you are using speakers vs headphones?

Regards

Quentin.

On Wed, Feb 26, 2020 at 6:38 AM Shaun Everiss <sm.everiss@...> wrote:

I have noticed with especially ends of words in espeak at any rate not sure with the fest but they seem to pop at the end of words and sound more abbrupt especially at the end of sentences.

It doesn't always happen but enough times a user would notice.

It doesn't mean the user can't use the software but you notice it when reading.



On 21/02/2020 5:02 pm, Quentin Christensen wrote:
We did rewrite a lot of the speech code, although for now this shouldn't be an overly different experience, it was more to make future changes and maintenance easier.  Evidently you've found something that does work slightly differently.

Not sure about the first one, I'll have to learn GMails keyboard shortcuts more - but re editing words, I've written that up https://github.com/nvaccess/nvda/issues/10808

Regards

Quentin.

On Fri, Feb 21, 2020 at 2:40 AM Giles Turnbull <giles.turnbull@...> wrote:
Hi all,

does anyone know whether there have been changes to the speech processing part of NVDA in the 2019.3 update? I didn't spot anything in the change log, but I experience a couple of things that are different from the 2019.2.1 version.

The first one is very minor ... when I tag an email in gmail so that it goes into an existing  folder, I press c to go to the combo box listing all folders, then I press a letter corresponding to the first letter of the folder name - such as f for my Facebook notifications. Once focus is on Facebook I tab over and activate the Go button, which applied the label to the email. Since 2019.3 the letter I pressed (in this case f) gets spoken when I tab over to the Go button; this did not happen before 2019.3.If I need to press the letter more than once to continue past Facebook to Farming and then to Fiction, NVDA now announces "fff" which is mildly annoying!

More of a problem is when I mistype words ... I regularly mistype because as becuase and words ending in tion as toin ... I suspect when I've done this and immediately select the word and retype it without presing space to move onto typing the next word. As a result the first mistyping doesn't get spoken straight away ... it gets spoken as I press space or punctuation ... and it merges the mistyped word in with the correct version, and they are seemlessly connected like they are one single word, becuasebecause ... the same happened in this post when I mistyped punctuation as puncutation and got "puncutationpunctuation" as a result.

The only irritation with this is that it makes it difficult to tell whether the subsequent attempts at spelling the word correctly has removed the typo or whether it is still mistyped.

I'm curious whether this is an intentional change and, if so, what the thinking behind it is.

Giles


--
Quentin Christensen
Training and Support Manager




--
Quentin Christensen
Training and Support Manager



 

Mine is just a standard realtech.

It only really does it when brousing the web in waterfox, no idea why but its random.

Clipping is probably a good way to describe it.

On 26/02/2020 1:22 pm, David Csercsics wrote:
I notice a similar issue with espeak where the beginning and end of utterances will sometimes clip with both of my USB sound cards. One of them is a AudioQuest DragonFly Red which is sometimes hooked to headphones and sometimes studio monitors. When I first start NVDA the first part of the startup tones gets clipped and the first syllable of the first utterance from espeak is clipped off as well. This happens less frequently with the Logitech h800 headphones I have, but even on those it still occurs occasionally.




 

So far I only notice it with nvda.



On 26/02/2020 1:26 pm, Quentin Christensen wrote:
I wonder does it happen with other sounds as well, eg the start of a music track, the first ring of Skype, etc?

My guess is it's likely the driver - allowing the device to try to play sounds before it has properly initialised etc, and I doubt there is too much NVDA itself can do.

On Wed, Feb 26, 2020 at 11:22 AM David Csercsics <bleeblat@...> wrote:
I notice a similar issue with espeak where the beginning and end of
utterances will sometimes clip with both of my USB sound cards. One of
them is a AudioQuest DragonFly Red which is sometimes hooked to
headphones and sometimes studio monitors. When I first start NVDA the
first part of the startup tones gets clipped and the first syllable of
the first utterance from espeak is clipped off as well. This happens
less frequently with the Logitech h800 headphones I have, but even on
those it still occurs occasionally.







--
Quentin Christensen
Training and Support Manager



 

Well lets see if it continues, I just got the latest driver with my card and its not doing it now and its a lot less allready.

So maybe eventually its going to go away.

I always get latest driver in fact things are a lot better now so who knows.

On 26/02/2020 1:44 pm, David Csercsics wrote:
No, I haven't seen it with other sounds unless the sound card was previously asleep for a long period. It's probably a driver bug somewhere. It's not too severe, just a bit strange. The sound card that glitches most frequently with NVDA only supports 24-bit audio, so maybe that has something to do with it.




David Csercsics
 

Well, both of these cards use the built-in Microsoft USB audio driver. I've got all the Windows audio enhancements disabled, because it seems that it is better for sound supposedly if you do that. I'm not sure what other implications that has though.


Gene
 

Use the silence utility and see if that helps.  It generates some sort of inaudible sound that keeps the sound card always on.  I may have it on my Dropbox and I'll send it later if I do.
 
Gene
----- Origianal Message -----

Sent: Tuesday, February 25, 2020 7:10 PM
Subject: Re: [nvda] changes in speech in NVDA 2019.3

Well, both of these cards use the built-in Microsoft USB audio driver.
I've got all the Windows audio enhancements disabled, because it seems
that it is better for sound supposedly if you do that. I'm not sure what
other implications that has though.



Felix G.
 

I don't think the speech refactor affects specific synths, so the
Espeak issues described here are probably the result of an Espeak
update or a subtle sound driver issue. I'm guessing though, so use
grain of salt heuristics
Greetings,
Felix

Am Mi., 26. Feb. 2020 um 17:35 Uhr schrieb Gene <gsasner@gmail.com>:


Use the silence utility and see if that helps. It generates some sort of inaudible sound that keeps the sound card always on. I may have it on my Dropbox and I'll send it later if I do.

Gene
----- Origianal Message -----
From: David Csercsics
Sent: Tuesday, February 25, 2020 7:10 PM
To: nvda@nvda.groups.io
Subject: Re: [nvda] changes in speech in NVDA 2019.3

Well, both of these cards use the built-in Microsoft USB audio driver.
I've got all the Windows audio enhancements disabled, because it seems
that it is better for sound supposedly if you do that. I'm not sure what
other implications that has though.