Date   
[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By rhdunn:

You are correct. I'll address that issue as well.

I have a fix for the en-GB-x-gbclan issue, which was in SubstitutePhonemes, not preserving the sourceix value for a phoneme that is replaced by NULL. I'm still trying to track down where the hyw issue occurs.

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

This is a related issue, but isn't the part in dictionary.c also a bug? It prints a space before the beginning of sentence, but I think it should only print a space between words. I mean, the way hyw and others using replace rules currently print seems logical, and the normal behavior of having an extra space seems odd.

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By rhdunn:

Thanks for investigating this. I have added constants to make newword easier to understand.

In phonemelist.c, sourceix is set to 10241 for hy and 0 for hyw. As such, it is not running the code that sets newword. This is the case whenever there are replace rules in the voice/language file, and that replace rule is used at the start of a word. For example:

$ ESPEAK_DATA_PATH=`pwd` src/espeak-ng -xqv en-GB-x-gbcwmd "have"
'av
$ ESPEAK_DATA_PATH=`pwd` src/espeak-ng -xqv en-GB-x-gbcwmd "I have"
 aI'av

Note how this is not just at the start of the sentence. This makes sense, because if sourceix is 0, then newword is 0, so the logic in dictionary.c to add a space on a new word is not run.

I'm not yet sure why sourceix is 0 in this case, but I suspect that the replacement phoneme is being copied, but the sourceix of the original phoneme is not being preserved.

Github push to espeak-ng:espeak-ng #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

1 New Commit:

[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
8e13f7147c19: Add constants for use with PHONEME_LIST.newword.

Modified: src/libespeak-ng/phonemelist.c
Modified: src/libespeak-ng/setlengths.c
Modified: src/libespeak-ng/synth_mbrola.c
Modified: src/libespeak-ng/synthesize.c
Modified: src/libespeak-ng/synthesize.h

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

Here's the code that prints a space before the first word when using espeak-ng -x: https://github.com/espeak-ng/espeak-ng/blob/d0e806b6009b863770f19254291d5749368ff78b/src/libespeak-ng/dictionary.c#L581-L582

plist.newword is used in a few places. I found this: https://github.com/espeak-ng/espeak-ng/blob/d0e806b6009b863770f19254291d5749368ff78b/src/libespeak-ng/phonemelist.c#L497-L506

For language hy, newword is set to 5, but for hyw it's 0. start_sentence is passed from translate.c where it's called new_sentence2, which is set by new_sentence, apparently here: https://github.com/espeak-ng/espeak-ng/blob/d0e806b6009b863770f19254291d5749368ff78b/src/libespeak-ng/translate.c#L2074-L2077

I can't figure out where the difference in new_sentence comes from.

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By rhdunn:

The language-phonemes.test scripts wrap the input in [[...]] (see line 30), so ma na pa is actually [[ma na pa]]. This tells espeak to use the text as phonemes, not as letters. Your output is based on reading text.

The command that gets run is things like espeak-ng -xv hyw --stdout "[[ma na pa]]" | sha1sum. This creates a SHA1 hash of the generated audio and compares it to the expected value.

I can confirm that the hashes are different, but don't know why. It may be related to the underlying issue, so a fix is therefore not straightforward :(. As such, it is better to hold off of creating a pull request until the phoneme difference can be identified.

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

My proposed fix would cause make check to fail on language-phonemes.test.

Here's the output of the first five phonemes. I don't know how to interpret the results. Is this a desired outcome, should I change the checksum in the test file so the test will pass?

Running espeak-ng -xv hyw "ma, na, Na, pa, p#a" Output before fix:

_:(en)'Em(hy) 'a
 _:(en)'En(hy) 'a
 _:(en)'En(hy) 'a
 _:(en)p'i:(hy) 'a
 _:(en)p'i: h'aS(hy) 'a

Output after fix:

(en)m'A:(hy)
 (en)n'A:(hy)
 (en)n'A:(hy)
 (en)p'A:(hy)
 (en)p'i: h'aS(hy) 'a

The other Armenian language produces this: espeak-ng -xv hy "ma, na, Na, pa, p#a"

 (en)m'A:(hy)
 (en)n'A:(hy)
 (en)n'A:(hy)
 (en)p'A:(hy)
 (en)p'i: h'aS(hy) 'a

Re: Medium of Reporting Mispronunciations

Reece H. Dunn
 

On Fri, Aug 17, 2018 at 08:28 PM, Bhavya shah wrote:
Dear all,

Often, I come across words that are simply mispronounced by ESpeak-NG.
To report these currently mispronounced words, should I file tickets
on GitHub or share a bunch of them every so often on this e-group?

I would appreciate any clarifications.
For reporting mispronunciations, either creating an issue/ticket on GitHub, or listing them here on this e-group is fine.

If reporting the mispronunciations to GitHub, it would be better to group mispronunciations for a language to reduce the number of issues. Likewise, if you find additional mispronunciations, adding a comment to an existing open issue would be helpful to further reduce the number of issues. If you do end up creating multiple issues because you cannot find a relevant open issue, that is okay as well.

Kind regards,
Reece

Thanks.

--
Best Regards
Bhavya Shah

Blogger at Hiking Across Horizons: https://bhavyashah125.wordpress.com/

Contacting Me
E-mail Address: bhavya.shah125@...
LinkedIn: https://www.linkedin.com/in/bhavyashah125/
Twitter: @BhavyaShah125
Skype: bhavya.09

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By rhdunn:

There are two separate issues. The first is the specific case described above with the hy-arevmda/hyw language, which your patch fixes.

The second is the root issue: why does the change to that language affect the replace rules (k => g) at the start such that it is not applied and the space is removed. That should be a separate issue, which can use this as a test case during development.

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

Do you think this is a proper fix for the root issue, no space before the phoneme output with -x? Also, I don't fully understand the effect of multiple language settings, their order and priority. This might break something else, pronunciation for certain words for example.

Re: Medium of Reporting Mispronunciations

Valdis Vitolins
 

The best way is to provide fixes with pull requests on
https://github.com/espeak-ng/espeak-ng/pulls

(Pull requests should pass checks and should not have conflicts.
If changes are only about spelling, you may create pull request
to my project https://github.com/valdisvi/espeak-ng to make life easier
for leading developer Reece H. Dunn.)

Second best way is to send updated configuration files in mailing list.
(Then usually I try to merge these files in my project and I create
pull request for Reece in espeak-ng main project.)

Third way is to provide list of words and describe, what should be
proper spelling as an issue in
https://github.com/espeak-ng/espeak-ng/issues

Single word may not be good candidate for issue, because usually
spelling should be described as more general rule instead of just
single word. (If word is in top 100..1000 of used words, it may have
its own spelling rule. All other words, if necessary, should be added
in specific lists in configuration files.)

In general, any notes, comments and contributions are considered.
Though, if handling of given information is more difficult,
it may take much more time.

Valdis

Dear all,

Often, I come across words that are simply mispronounced by ESpeak-
NG.
To report these currently mispronounced words, should I file tickets
on GitHub or share a bunch of them every so often on this e-group?

I would appreciate any clarifications.

Thanks.

Re: odd behavior with the word whereever

Travis Siegel
 

No, each time it happens, it's when reading via the autoread function of NVDA (NVDA-Down arrow), and only when it's in that mode. If I readby line, it doesn't happen, and it doesn't happen if I read by word, so I can't for the life of me figure out what is causing it to do that. Again, it's running at 99 percent, I suppose the next time it comes up, I can try lowering the speed, and see if it still does it. I'll see what I can figure out about what triggers it, and what doesn't. Everytime it's hapened though, the surrounding text hasn't been the same, so it's got to be the words themselves.

On Wed, 15 Aug 2018, Kyle wrote:

Very odd indeed. Are the surrounding words the same or similar every time? What if either whatever or wherever is found on a line by itself, although that in itself is rather rare. I can't confirm it here, so it seems like the problem is related specifically to NVDA somehow. I tried running espeak-ng from a shell and I tried running it with Orca/speech-dispatcher, and none of these produces this behavior. No luck trying in a sentence or standalone.
Imetumwa kutoka kitunguu changu


Medium of Reporting Mispronunciations

Bhavya shah
 

Dear all,

Often, I come across words that are simply mispronounced by ESpeak-NG.
To report these currently mispronounced words, should I file tickets
on GitHub or share a bunch of them every so often on this e-group?

I would appreciate any clarifications.

Thanks.

--
Best Regards
Bhavya Shah

Blogger at Hiking Across Horizons: https://bhavyashah125.wordpress.com/

Contacting Me
E-mail Address: bhavya.shah125@...
LinkedIn: https://www.linkedin.com/in/bhavyashah125/
Twitter: @BhavyaShah125
Skype: bhavya.09

Re: odd behavior with the word whereever

Bhavya shah
 

Hi Travis,

I have also experienced this very oddity you have described in recent
times. I have never however attempted to drill down on when this
strange speech occurs and when it does not. I would also add that
sometimes, there one ever or something similar sounding is added to
wherever, making it sound like something along the lines of
wherevereeevereevereevereeveree. If and when I encounter this again, I
will try to investigate this issue and report back with findings and
reliable test cases on this list.

Thanks.

P.S. I could not replicate this bug while reading your e-mails.

On 8/16/18, Kyle <kyle@...> wrote:
Very odd indeed. Are the surrounding words the same or similar every
time? What if either whatever or wherever is found on a line by itself,
although that in itself is rather rare. I can't confirm it here, so it
seems like the problem is related specifically to NVDA somehow. I tried
running espeak-ng from a shell and I tried running it with
Orca/speech-dispatcher, and none of these produces this behavior. No
luck trying in a sentence or standalone.
Imetumwa kutoka kitunguu changu



--
Best Regards
Bhavya Shah

Blogger at Hiking Across Horizons: https://bhavyashah125.wordpress.com/

Contacting Me
E-mail Address: bhavya.shah125@...
LinkedIn: https://www.linkedin.com/in/bhavyashah125/
Twitter: @BhavyaShah125
Skype: bhavya.09

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By rhdunn:

Thanks for the fix. Can you create a pull request or a branch on your repository so you are correctly credited for the fix?

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

The patch below would fix this particular case. I've yet to figure out why changing the order of language parameters has an effect on anything (EDIT: something to do with V_LANGUAGE in voices.c).

diff --git a/espeak-ng-data/lang/ine/hyw b/espeak-ng-data/lang/ine/hyw
index 9d5fe86..48516cd 100644
--- a/espeak-ng-data/lang/ine/hyw
+++ b/espeak-ng-data/lang/ine/hyw
@@ -1,7 +1,7 @@
 name Armenian (West Armenia)
-language hyw
 language hy-arevmda
 language hy  8
+language hyw
 
 dictionary hy
 dictrules 1

[espeak-ng:master] reported: Tests: strange output using language variant #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Issue #528 Tests: strange output using language variant
By jaacoppi:

The patch below would fix this particular case. I've yet to figure out why changing the order of language parameters has an effect on anything.

diff --git a/espeak-ng-data/lang/ine/hyw b/espeak-ng-data/lang/ine/hyw
index 9d5fe86..48516cd 100644
--- a/espeak-ng-data/lang/ine/hyw
+++ b/espeak-ng-data/lang/ine/hyw
@@ -1,7 +1,7 @@
 name Armenian (West Armenia)
-language hyw
 language hy-arevmda
 language hy  8
+language hyw
 
 dictionary hy
 dictrules 1

Re: odd behavior with the word whereever

Kyle <kyle@...>
 

Very odd indeed. Are the surrounding words the same or similar every time? What if either whatever or wherever is found on a line by itself, although that in itself is rather rare. I can't confirm it here, so it seems like the problem is related specifically to NVDA somehow. I tried running espeak-ng from a shell and I tried running it with Orca/speech-dispatcher, and none of these produces this behavior. No luck trying in a sentence or standalone.
Imetumwa kutoka kitunguu changu

odd behavior with the word whereever

Travis Siegel
 

I've been noticing lately, that when NVDA comes across the word whatever or whereever, for some reason, it works just fine if I'm reading by line, or check it by word, but if it's reading via the NVDA-Down Arrow, when it comes across the wors whatever or whereever, it adds an extra ever after it, but the first E is a long e, so it sounds like it's saying where ever eever.
I've confirmed this multiple times, but I can't for the life of me figure out what would trigger this, since reading line by line or checking the word doesn't do this, only when it's reading down the screen automatically. I'm using the ivan 3 voice, and I'm running it at 99 percent, but I don't know why the speed would affect how it pronounces things. Very odd.
Anyway, just figured I'd report here, in case someone else has an idea on what might be causing it, or how to fix it.

[espeak-ng:master] reported: Add flag to flush stdout #github

espeak-ng@groups.io Integration <espeak-ng@...>
 

[espeak-ng:master] New Comment on Pull Request #536 Add flag to flush stdout
By micha137:

@rhdunn Thank you for the review. I will amend the pull request accordingly.