Date   

Updates to Github #github

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

[espeak-ng:master] New Issue Created by rhdunn:
#271 Fix building emoji for ml, my, ne, pa and ta

The following languages cause a segfault when compiling/using the dictionary with emoji support added to them:

  • [x] ml (Malayalam)
  • [ ] my (Myanmar/Burmese)
  • [ ] ne (Nepali)
  • [x] pa (Punjab)
  • [ ] ta (Tamil)


[espeak-ng:master] New Issue Created by rhdunn:
#271 Fix building emoji for ml, my, ne, pa and ta

The following languages cause a segfault when compiling/using the dictionary with emoji support added to them:

  • [x] ml (Malayalam)
  • [x] my (Myanmar/Burmese)
  • [ ] ne (Nepali)
  • [x] pa (Punjab)
  • [ ] ta (Tamil)


[espeak-ng:master] New Issue Created by rhdunn:
#271 Fix building emoji for ml, my, ne, pa and ta

The following languages cause a segfault when compiling/using the dictionary with emoji support added to them:

  • [x] ml (Malayalam)
  • [x] my (Myanmar/Burmese)
  • [x] ne (Nepali)
  • [x] pa (Punjab)
  • [ ] ta (Tamil)


6 New Commits:

[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
cec0e773a403: Remove the tokenizer code. This is not currently used, and should be designed from espeak's behaviour.

Removed: src/include/espeak-ng/tokenizer.h
Removed: src/libespeak-ng/tokenizer.c
Removed: tests/tokenizer.c
Modified: Makefile.am
Modified: android/jni/Android.mk
Modified: src/libespeak-ng/readclause.c
Modified: src/windows/libespeak-ng.vcxproj
Modified: src/windows/libespeak-ng.vcxproj.filters


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
0bb4501253fc: Revert "pa: remove the emoji support for now, as it fails to build" This reverts commit 593949dec62d6cdc78ed8383e113c48654829243.

Added: dictsource/pa_emoji
Modified: CHANGELOG.md
Modified: Makefile.am


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
2642ba8ed5f4: ml: (emoji) -- CLDR 31.0.1 emoji translations

Added: dictsource/ml_emoji
Modified: CHANGELOG.md
Modified: Makefile.am


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
3bc6da0d0725: my: (emoji) -- CLDR 31.0.1 emoji translations

Added: dictsource/my_emoji
Modified: CHANGELOG.md
Modified: Makefile.am


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
bc7d806505ca: ne: (emoji) -- CLDR 31.0.1 emoji translations

Added: dictsource/ne_emoji
Modified: CHANGELOG.md
Modified: Makefile.am


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
8d07d2eacd17: ta: (emoji) -- CLDR 31.0.1 emoji translations

Added: dictsource/ta_emoji
Modified: CHANGELOG.md
Modified: Makefile.am


[espeak-ng:master] New Issue Created by rhdunn:
#271 Fix building emoji for ml, my, ne, pa and ta

The following languages cause a segfault when compiling/using the dictionary with emoji support added to them:

  • [x] ml (Malayalam)
  • [x] my (Myanmar/Burmese)
  • [x] ne (Nepali)
  • [x] pa (Punjab)
  • [x] ta (Tamil)


[espeak-ng:master] new issue: Fix building emoji for ml, my, ne, pa and ta #github

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

[espeak-ng:master] New Issue Created by rhdunn:
#271 Fix building emoji for ml, my, ne, pa and ta

The following languages cause a segfault when compiling/using the dictionary with emoji support added to them:

  • [ ] ml (Malayalam)
  • [ ] my (Myanmar/Burmese)
  • [ ] ne (Nepali)
  • [x] pa (Punjab)
  • [ ] ta (Tamil)


[espeak-ng:master] new issue: New espeak variant: Pablo #github

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

[espeak-ng:master] New Issue Created by ehollig:
#298 New espeak variant: Pablo

Forwarding this on from https://github.com/nvaccess/nvda/issues/2704. pablo.zip


[espeak-ng:master] reported: en : velarization of /n/ to /N/ before velar stops might have a too large scope #github

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

[espeak-ng:master] New Comment on Issue #297 en : velarization of /n/ to /N/ before velar stops might have a too large scope
By rhdunn:

Regarding the pronunciations, they are derived as an extension of John Wells' Lexical Sets (see https://github.com/espeak-ng/espeak-ng/blob/master/docs/languages/gmw/en.md). These are sometimes customised specifically for different accents (e.g. American variants of words like vase and antibody) -- which is why you will see things like ?3 and ?!5 for various rules and words. I want to make this configuration more flexible as documented in #286 to allow for different accent configurations. The specific rendering of those phonemes then varies across the different English accents.

My understanding of velar assimilation of /n/ before /k/ and /g/ (specifically within a word) is that it is universal in English accents -- unless you have a specific case where this is not true. Dictionaries are an unreliable source as they can and do omit phonological features (like aspiration of /p/, and other stops at the start of a word/syllable). NOTE 1: There is variation of -ng words (ringer as /rINg3/ or /rIN3/), and -ing (e.g. adding as /adINg/, /adIN/ or /adIn/ -- note that the latter is due to the lack of the /g/ influencing the velarization of /n/). NOTE 2: It is difficult to pronounce the /n/ without velarization when the /n/ is followed by /k/ or /g/, compared with the other examples on that page (with the exception of apprenticeship, where the /sS/ is also difficult to pronounce). The -in' variant of -ing is easy because in that case the /g/ sound is not pronounced.

I agree that the other features vary with accent, formality (whether or not you are carefully enunciating the words) and how fast you are speaking. I have not looked into these yet and where best to apply them.


[espeak-ng:master] reported: en : velarization of /n/ to /N/ before velar stops might have a too large scope #github

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

[espeak-ng:master] New Comment on Issue #297 en : velarization of /n/ to /N/ before velar stops might have a too large scope
By BenTalagan:

Hello Reece and thanks a lot for taking some of your time to answer in detail to this question!

Thanks also for highlighting the difference of approach between broad and narrow transcriptions ; that point of theory makes things much clearer for me concerning the motivations of phoneticians.

Though the velar assimilation process in that context is pretty clear to me, there are one or two points on which I'd like to humbly ask for some clarification. The degree of assimilation may vary a lot depending of the accent, dialect, or even the individual way of speaking of a locutor. This degree of assimilation is somewhat part of what constitutes a continuum between broad and narrow transcriptions, since a phonetical colour will happen to be rendered (or not) depending on where you place the cursor for the level of detail you want for the rendering. Of course, for a text to speech application such as espeak, the reasons for choosing a narrower transcription is perfectly understandable, even obvious. For a dictionary, I guess the motivations are not exactly the same (phonology, morphology, influence of the etymology & so on).

In the way espeak is designed, all english prononciations are derived from a common prononciation which is "English_(Great_Britain)" ; this means that particular phonetic colors and details brought to this common trunk will be inherited by other prononciations (unless they are canceled by override in the derivate files). It seems a bit odd to me since I would have expected to have a more "neutral", broader form of english with less colors as the trunk (even if would be a bit artificial), from which the GB English prononciation would have derived just like the other ones, to avoid this inheritance (but maybe I'm wrong and it can be justified by some reality).

On that specific point of velar assimilation, it seems to me (but I can be misleaded since I am not a native english speaker) that the degree of assimilation for certain specific prononciations such as the conservative RP which is very articulated (and somewhat 'perseverative') would be so low that choosing to render it with ŋ would tend to over-highlight it compared to reality.

This question of degree does puzzles me all the more that other assimilations that may exist in english are not rendered in espeak : for exemple, the labial assimilation of n in m before p and b (which is very common in french and appears in the orthograph itself) may also be present in english (cf espeak rendering of input ˈɪnpʊt and inbox ˈɪnbɒks, which could have been, as well, ˈɪmpʊt and ˈɪmbɒks ). So what is the reason behind these choices, is it really a question of quantity and cursor, as I suspect ?

Thanks!

Ben


Updates to Github #github

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

4 New Commits:

[espeak-ng:master] By chrislm <llajta2012@...>:
4c696620bb93: IT: new improvements (tested on june/july 2017)

Modified: dictsource/it_list
Modified: dictsource/it_listx
Modified: dictsource/it_rules


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
afa99784ab9a: en: various pronunciation fixes

Modified: dictsource/en_list
Modified: dictsource/en_rules


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
2081a51ff8fe: Merge remote-tracking branch 'Christianlm/master'

Modified: dictsource/it_list
Modified: dictsource/it_listx
Modified: dictsource/it_rules


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
9ac98311294d: af: Updates from Christo de Klerk.

Modified: dictsource/af_rules


[espeak-ng/espeak-ng] Pull request closed by rhdunn:

#295 IT: new improvements (tested on june/july 2017)

Pronunciation fixes for Italian language.


Update rules file for Afrikaans

Christo de Klerk
 

Hello Reece


I have made a few pronunciation corrections in af_rules and have attached it here.


Many thanks for the help.


Kind regards


Christo


Updates to Github #github

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

[espeak-ng:master] New Comment on Issue #297 en : velarization of /n/ to /N/ before velar stops might have a too large scope
By rhdunn:

This is correct. It is intended to apply to things like the in- and un- prefices. Dictionaries (and other broad transcriptions) leave it out because /N/ is an allophone of /n/ (i.e. /n/ can be /N/ in some circumstances). This is like how stops like /p/ are aspirated at the start of a word (e.g. spot, pot), but that aspiration is not transcribed. Narrow transcriptions preserve these features.

English is a complex case because /N/ is an allophone (e.g. incoherent) and a phoneme (e.g. king). This is because in king and similar words, the /g/ is dropped (elided), but the /n/ is still velarized (i.e. /kIng/ => /kINg/ => /kIN/).

This change of /n/ to /N/ before /k/ and /g/ is a process called velar assimilation [2], [3], or velarization (assimilation affecting the velar place of articulation):

If you sound out words like incoherent and uncouth slowly, you will notice that when pronouncing the first /n/, the body of the tongue is touching the velar ridge (the large cavity behind the gum, before the tonsils) [1]. Here, the tongue is moving to the velar ridge to provide the stop part of the /k/ or /g/ sound, before the release part of those stops. If you compare this with words like intolerable and untied, the tip of the tongue touches the flat part of the gum behind the teeth, just before it moves upward (the alveolar ridge) [1].

If anything, it should only apply across words where there are no pauses (i.e. clause breaks).

[1] https://en.wikipedia.org/wiki/Place_of_articulation [2] https://en.wikipedia.org/wiki/Assimilation_(phonology) [3] http://martinweisser.org/courses/phonetics/connect/assimilation.html


[espeak-ng:master] Issue #297 en : velarization of /n/ to /N/ before velar stops might have a too large scope closed by BenTalagan.


[espeak-ng:master] Label added to issue #297 en : velarization of /n/ to /N/ before velar stops might have a too large scope by BenTalagan.


[espeak-ng:master] new issue: en : velarization of /n/ to /N/ before velar stops might have a too large scope #github

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

[espeak-ng:master] New Issue Created by BenTalagan:
#297 en : velarization of /n/ to /N/ before velar stops might have a too large scope

Hello,

On this commit 6a85d3e63242411bbe5ef9b439805c43b07354d2 has been introduced the following phoneme rule for general english :

phoneme n
  vcd alv nas
  ipa n
  IF nextPh(isVelar) THEN // /n/ is velarized to /N/ when before velar stops
    ChangePhoneme(N)
  ENDIF
  CALL base1/n
endphoneme

But isn't the scope of that rule a bit too large ? It would affect all english voices/accents/pronunciations on inter-word links such as

in Glasgow > ɪŋ ɡlˈazɡəʊ

but also affect the pronunciation of in- and un- prefixes. Cf :

https://www.merriam-webster.com/dictionary/incoherent https//www.merriam-webster.com/dictionary/uncouth


[espeak-ng:master] reported: Pronunciation of 'reputedly' is broken in current master for various english langs #github

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


[espeak-ng:master] reported: Reimplement the SAPI bindings. #github

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

[espeak-ng:master] New Comment on Issue #7 Reimplement the SAPI bindings.
By rhdunn:

Open espeak SAPI bug: https://github.com/nvaccess/nvda/issues/1736.


Updates to Github #github

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

[espeak-ng:master] New Issue Created by rhdunn:
#296 The Russian voice speaks faster than the English one

Running:

$ time espeak-ng -vru "Александр Македонский"
real    0m1.508s
$ time espeak-ng -vru "Аleksander Makidonski"
real    0m1.610s
$ time espeak-ng -vbg "Александр Македонский"
real    0m1.670s

The Russian version is perceptably faster than the English and Bulgarian versions. This is especially true when Russian and English are mixed. NOTE: A +/- 0.1s difference seems to be the threshold as the Bulgarian is not (to me) perceptably slower when compared to the English (a +/- 0.06s difference).

Original issue: https://github.com/nvaccess/nvda/issues/2441


[espeak-ng:master] Label added to issue #296 The Russian voice speaks faster than the English one by rhdunn.


[espeak-ng:master] Label added to issue #296 The Russian voice speaks faster than the English one by rhdunn.


Updates to Github #github

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


[espeak-ng:master] reported: Pronunciation of 'reputedly' is broken in current master for various english langs #github

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

[espeak-ng:master] New Comment on Issue #294 Pronunciation of 'reputedly' is broken in current master for various english langs
By BenTalagan:

(Seems solved by commit 01919cd48a566cdf34347784b2e74554b376e900 ? , right ? 😌 )


Pull Request Opened #github

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

[espeak-ng/espeak-ng] Pull request opened by Christianlm:

#295 IT: new improvements (tested on june/july 2017)

Pronunciation fixes for Italian language.


[espeak-ng:master] reported: Can I contribute if I am not a programmer? #github

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

[espeak-ng:master] New Comment on Issue #291 Can I contribute if I am not a programmer?
By orschiro:

Thanks! Coming back to one point:

eSpeakNG requires specific set of technical skills to improve it, therefore list of its contributors is not long. But, if there would be considerable interest, we could create e.g. Patreon page to pay money for its contributors.

I think this is an excellent idea! I feel it's quite hard for me to contribute here but would love to give you a pledge to help espeak on Linux. :-)


Updates to Github #github

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

[espeak-ng:master] New Comment on Issue #291 Can I contribute if I am not a programmer?
By valdisvi:

Yes. Actual executable for eSpeak NG is called espeak-ng, and for compatibility reasons it also has symbolic link named just espeak. (You will get this if you will install espeak-ng from source code, but if you will install binary package from repository, it depends how package maintainers decided to deal with this.)


[espeak-ng:master] New Comment on Issue #291 Can I contribute if I am not a programmer?
By valdisvi:

Yes. Actual executable for eSpeak NG is called espeak-ng, and for compatibility reasons it also has symbolic link named just espeak. (You will get this if you will install espeak-ng from source code, but if you will install binary package from repository, it depends how package maintainers decided to deal with this.)


[espeak-ng:master] reported: Can I contribute if I am not a programmer? #github

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

[espeak-ng:master] New Comment on Issue #291 Can I contribute if I am not a programmer?
By valdisvi:

Thanks for your interest!

Anybody can contribute to pronunciation rules for language, which he or she knows (e.g. German, Polish, English etc.). You can submit just issues about incorrect pronunciation in its phonetic notation, e.g. "car" should be said as "k'A@".

But, better would be considering translation rules, which are described in dictionary.md file. Configuration files for e.g. English are: en_rules and en_list. You can test quite recent development version of espeak-ng online page.

There are some open issues about documentation and languages, but these are quite technical and require understanding of espeak-ng infrastructure.

eSpeakNG uses formant synthesis to produce sound, therefore recorded voice can be used only to get voice spectrum, which then can be approximately emulated by eSpeakNG. There are other open source synthesizers, which use recorded voice, e.g. Festival and closed source like MBROLA, but also for these, making of recordings is just first step in data analysis process of particular language and voice.

eSpeakNG requires specific set of technical skills to improve it, therefore list of its contributors is not long. But, if there would be considerable interest, we could create e.g. Patreon page to pay money for its contributors.


Updates to Github #github

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

[espeak-ng:master] New Comment on Issue #287 Buffer overflow when compiling dictionaries
By feerrenrut:

Thanks for looking at this so quickly @rhdunn. This seems to resolve the issue that we were seeing.


[espeak-ng:master] New Comment on Issue #290 Use proper structures to represent the hash chain used in dictionaries.
By feerrenrut:

At the risk of stating something which might be obvious. Don't forget to handle differences in endian-ness in the binary storage of the "length" parameter.


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@...>:
01919cd48a56: en: fix -edly pronunciation.

Modified: dictsource/en_rules

3801 - 3820 of 4658