Updates to Github #github


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

[espeak-ng:master] New Comment on Issue #933 Unable to compile zhy dictionay on Windows
By feerrenrut:

Yes. thanks @jaacoppi. We are updating NVDA to make use of this in nvaccess/nvda#12370


[espeak-ng:master] Issue #933 Unable to compile zhy dictionay on Windows closed by feerrenrut.


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

[espeak-ng:master] New Comment on Issue #950 eSpeak does not restore the language after speaking particular characters that forceably use a different language
By michaelDCurran:

Apologies. I did not clearly test using the espeak-ng executable carefully enough it seems. Agreed that the testcase I gave does not change. However, reverting that commit from eSpeak most certainly resolves the issue in NVDA. Therefore I can at least be sure that with that commit, the language change caused by that character continues during subsequent calls to espeak_Synth. Where as if that commit is reverted, the language is correctly reset at or before the end of the espeak_Synth call containing that character, and further espeak_Synth calls use the original language as expected.


[espeak-ng:master] New Comment on Issue #950 eSpeak does not restore the language after speaking particular characters that forceably use a different language
By michaelDCurran:

I'm not too sure now how to reproduce this with the espeak-ng executable. What would be an acceptable testcase for proving the bug using the espeak_Synth API call?


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

1 New Commit:

[espeak-ng:master] By ronaldaug <saisenfa@...>:
1df852a1ae8c: add 'aWi' into vowels group

Modified: dictsource/shn_rules


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

#952 add 'aWi' into vowels group

add 'aWi' into vowels group


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

Sorry, I don't know programming myself and I'm blind, so it's hard for me to do that. Wish you sympathize.


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By valdisvi:

Simple feedback should be doable for anybody, who wants to improve eSpeak NG. If you find some accessibility issue, which makes your contribution hard, you are welcome to report about this as well.


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

[espeak-ng:master] New Comment on Issue #953 Shan Language tone number doesn't show up correctly.
By ronaldaug:

I'm not experienced with tonal languages. Maybe you can compare Shan with e.g. Chinese settings and try tr->langopts.tone_numbers = 1; or some other settings. Maybe @michael-conrad may help you, because he made tones from scratch for Cherokee language.

Thanks for quick reply, I forget to mention that all 6 tones' pitch and sound are working fine. (I can confirm that all the numbers define in shn_list sound correctly except the rules shn_rules ). The problem is, It doesn't follow in a correct order. For instance, if I type လၢမ်း, normally it should output la:m4 but currently it outputs l'a:1m. Some tones are not following as the last character. Anyway, thanks for the help, I will learn Chinese or Vietnam rules first then continuous with this issue again.


Here's some more ideas for debugging:

  1. You are getting "m'a1" because the code sets primary stress. This shouldn't happen with tone languages, it should be "ma1". See stress related options in tr_languages.c and compare them to chinese.

  2. Debug with only one word at a time so the tone sandhi in CalcPitches_Tone() isn't affecting the words.

Check dictsource/cmn_rules. The tones there are 55, 35, 214 and so on. Check CalcPitches_Tone() in intonation.c. There's code that handles those tones.

  1. use -X parameter instead of -x to see which rules are selected.

Thanks @jaacoppi , I will try to check dictsource/cmn_rules.


[espeak-ng:master] New Comment on Issue #953 Shan Language tone number doesn't show up correctly.
By ronaldaug:

I'm not experienced with tonal languages. Maybe you can compare Shan with e.g. Chinese settings and try tr->langopts.tone_numbers = 1; or some other settings. Maybe @michael-conrad may help you, because he made tones from scratch for Cherokee language.

Thanks for quick reply, I forget to mention that all 6 tones' pitch and sound are working fine. (I can confirm that all the numbers define in shn_list sound correctly except the rules shn_rules ). The problem is, It doesn't follow in a correct order. For instance, if I type လၢမ်း, normally it should output la:m4 but currently it outputs l'a:1m. Some tones are not following as the last character. Anyway, thanks for the help, I will learn Chinese or Vietnam rules first then continuous with this issue again.


Here's some more ideas for debugging:

  1. You are getting "m'a1" because the code sets primary stress. This shouldn't happen with tone languages, it should be "ma1". See stress related options in tr_languages.c and compare them to chinese.

  2. Debug with only one word at a time so the tone sandhi in CalcPitches_Tone() isn't affecting the words.

Check dictsource/cmn_rules. The tones there are 55, 35, 214 and so on. Check CalcPitches_Tone() in intonation.c. There's code that handles those tones.

  1. use -X parameter instead of -x to see which rules are selected.

Thanks @jaacoppi , I will try to check dictsource/cmn_rules.


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

[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

Can I record and send it? How to fill out the report? I will give you a few Vietnamese speaking websites for your reference.


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

You can learn Vietnamese by yourself through the Vietnamese community in your country.


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

How can I report bugs when I can't code?


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

[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

I will send you some articles by voice so you can refer to their pronunciation: Northern accent: https://www.youtube.com/watch?v=KS8bQertS_s Central accent: https://www.youtube.com/watch?v=16mcAmsxiXg Southern accent: https://thuviensachnoihuongduong.com/khiemthi/26/392 I think, you should refer to the pronunciation of it before making a decision to improve it. I recommend that people be able to record their voice and use AI to analyze it, then let the machine learn and come up with the most accurate pronunciation.


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

I will send you the alphabet pronunciation file later.


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

[espeak-ng:master] New Comment on Issue #954 Espeak takes too long to phonemize on colab
By Jaskaran170599:

Hi ,I'm training for TTS model and want to convert my hindi sentences into phonemized version using espeak-ng. Im using a python package called Phonemizer.

How do I install version 1.50 on colab?


[espeak-ng:master] New Comment on Issue #954 Espeak takes too long to phonemize on colab
By Jaskaran170599:

Hi @jaacoppi ,I'm training for TTS model and want to convert my hindi sentences into phonemized version using espeak-ng. Im using a python package called Phonemizer.

How do I install version 1.50 on colab?


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

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

#955 Issue 945

This is the "dynamic allocation" fix. Note that that bug isn't in the phoneme_list handling, it is in ph_list2 handling. The circular buffer fix might have worked but lacked attraction given the number of globals. This is a step to removing all the globals are therefore allowing creation of an MT capable libespeak.


[espeak-ng:master] New Comment on Issue #945 The change to tests/translate.test to detect #824 causes unexplained failures on some builds
By jbowler:

Fixed by the pull requests, if adopted.


[espeak-ng:master] Issue #945 The change to tests/translate.test to detect #824 causes unexplained failures on some builds closed by jbowler.


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

[espeak-ng:master] New Comment on Pull Request #955 Issue 945
By jbowler:

By "last commit" you mean 2d3cc580b8f837aabad369665f677846490ccdbd, right?

This has to be a single commit; it makes all the allocations after the ReadClause call up to but not including the MakePhonemeList call in TranslateClause dynamic. It wouldn't even compile if it were split up let alone pass "make check".

I don't care at all about the comments; they were just to help me because I couldn't readily see which pointer arguments were input, output or in-out. The "const" on the global phoneme_tab is part of the "constify" commit 13f5039491a6380c9e90febff5ffb90e0b0868a1; it seems useful to me but it doesn't matter to me. Removing the array sizes is part of isolating N_PHONEME_LIST, but that's part of the next change (to allocate phoneme_list and ph_list3 dynamically).

I've noticed -- especially in 2d3cc58 -- that there are a lot of extraneous whitespace changes, changing spaces to tabs in comments. The general rule in the espeak-ng codebase is "tabs to indent, spaces to align" so tabs should only appear at the start of lines and not in comments.

My default vi configuration is space indent, which is why this started happening. I changed to this:

:set noexpandtab :set copyindent :set preserveindent :set softtabstop=0 :set shiftwidth=4 :set tabstop=4

Is that correct? It comes from this URL: https://vim.fandom.com/wiki/Indent_with_tabs,_align_with_spaces

When I discovered what was going on I tried to re-indent everything correctly, but it looks like the above is allow tab indentation after the first non-tab too... It would be good to work out the correct settings and create an _vimrc in the project (the directories with source files) to keep it right. I'll do a re-indent commit when I've fully verified all the non-space changes.


[espeak-ng:master] New Comment on Pull Request #955 Issue 945
By jbowler:

I've noticed -- especially in 2d3cc58 -- that there are a lot of extraneous whitespace changes, changing spaces to tabs in comments. The general rule in the espeak-ng codebase is "tabs to indent, spaces to align" so tabs should only appear at the start of lines and not in comments.

My default vi configuration is space indent, which is why this started happening. I changed to this:

:set noexpandtab :set copyindent :set preserveindent :set softtabstop=0 :set shiftwidth=4 :set tabstop=4

Is that correct? It comes from this URL: https://vim.fandom.com/wiki/Indent_with_tabs,_align_with_spaces

When I discovered what was going on I tried to re-indent everything correctly, but it looks like the above is allow tab indentation after the first non-tab too... It would be good to work out the correct settings and create an _vimrc in the project (the directories with source files) to keep it right. I'll do a re-indent commit when I've fully verified all the non-space changes.


[espeak-ng:master] New Comment on Pull Request #955 Issue 945
By jbowler:

I've noticed -- especially in 2d3cc58 -- that there are a lot of extraneous whitespace changes, changing spaces to tabs in comments. The general rule in the espeak-ng codebase is "tabs to indent, spaces to align" so tabs should only appear at the start of lines and not in comments.

My default vi configuration is space indent, which is why this started happening. I changed to this:

:set noexpandtab :set copyindent :set preserveindent :set softtabstop=0 :set shiftwidth=4 :set tabstop=4

Is that correct? It comes from this URL: https://vim.fandom.com/wiki/Indent_with_tabs,_align_with_spaces

When I discovered what was going on I tried to re-indent everything correctly, but it looks like the above allows tab indentation after the first non-tab too; I didn't think about that... It would be good to work out the correct settings and create an _vimrc in the project (the directories with source files) to keep it right. I'll do a re-indent commit when I've fully verified all the non-space changes.


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

[espeak-ng:master] New Comment on Pull Request #955 Issue 945
By jbowler:

I'm having a hard time understanding the last commit, as it is making a lot of changes and I cannot wrap my head around what has changed from looking at the diff only. Having that broken up into separate commits would make it more manageable to review. Also, each commit is doing multiple things, which makes it harder for someone looking at the code (especially if a git bisect identifies one of these commits as breaking something) to verify that it is correct and if not what is wrong with it.

By "last commit" you mean 2d3cc580b8f837aabad369665f677846490ccdbd, right?

This has to be a single commit; it makes all the allocations after the ReadClause call up to but not including the MakePhonemeList call in TranslateClause dynamic. It wouldn't even compile if it were split up let alone pass "make check".

Because of the prior commits it ends up being a highly localized change - just translate.c. Similar changes are possible in readclause.c (for source[N_TR_SOURCE], as was) and makephonemelist.c (for phoneme_list, ph_list3) though the latter may not be necessary as the length of ph_list2 is already known.

One major problem is that much of the language handling is encoded in the C code including code in dictionary.c that is used for parsing. So language developers will change code that itself has to be changed significantly for dynamic allocation. My feeling is that forking a new major version will effectively disconnect that version from language improvements.


[espeak-ng/espeak-ng] Pull request updated by jbowler:

#955 Issue 945

This is the "dynamic allocation" fix. Note that that bug isn't in the phoneme_list handling, it is in ph_list2 handling. The circular buffer fix might have worked but lacked attraction given the number of globals. This is a step to removing all the globals are therefore allowing creation of an MT capable libespeak.


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

[espeak-ng/espeak-ng] Pull request updated by jbowler:

#955 Issue 945

This is the "dynamic allocation" fix. Note that that bug isn't in the phoneme_list handling, it is in ph_list2 handling. The circular buffer fix might have worked but lacked attraction given the number of globals. This is a step to removing all the globals are therefore allowing creation of an MT capable libespeak.


6 New Commits:

[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
5791bf28de91: Fix -Wunused-parameter warnings.

Modified: src/libespeak-ng/speech.c
Modified: tests/ssml-fuzzer.c


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
ed6e096b5b5b: Fix -Wunused-function warnings.

Modified: src/libespeak-ng/sPlayer.c


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
53283a910aa5: Fix -Wunused-but-set-variable warnings.

Modified: src/libespeak-ng/wavegen.c


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
720405c5c602: Fix -Wreturn-type warnings.

Modified: src/libespeak-ng/wavegen.c


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
f581436deb2d: Fix -Wmissing-prototypes warnings.

Modified: src/libespeak-ng/sPlayer.h


[espeak-ng:master] By Reece H. Dunn <msclrhd@...>:
a13f9bb62a45: Fix -Wint-conversion warnings.

Modified: configure.ac
Modified: src/libespeak-ng/synthesize.c


[espeak-ng:master] New Comment on Pull Request #955 Issue 945
By jbowler:

That should have removed all the space borkage. There are a lot of [^^I]^I sequences in there - a lot of cases where tabs are used within comments, but those aren't ones that I added which is probably why when I tried to re-indent translate.c and phonemelist.c things got changed.


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

[espeak-ng/espeak-ng] Pull request updated by jbowler:

#955 Issue 945

This is the "dynamic allocation" fix. Note that that bug isn't in the phoneme_list handling, it is in ph_list2 handling. The circular buffer fix might have worked but lacked attraction given the number of globals. This is a step to removing all the globals are therefore allowing creation of an MT capable libespeak.


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

#955 Issue 945

This is the "dynamic allocation" fix. Note that that bug isn't in the phoneme_list handling, it is in ph_list2 handling. The circular buffer fix might have worked but lacked attraction given the number of globals. This is a step to removing all the globals are therefore allowing creation of an MT capable libespeak.


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

[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

â, ă, ô, ơ, u, ư, ê, đ, d. Those are the words that need to be corrected in pronunciation. You have to know Vietnamese then when I speak you can understand the standard pronunciation, but if I speak vocabulary, then in essence, if I speak English, it will be different from the Vietnamese I told my friends.


[espeak-ng:master] New Comment on Issue #951 espeak has not been pronounced properly with other languages
By NVCDevelopmentTeam:

â, ă, ô, ơ, ư, ê, đ, d. Those are the words that need to be corrected in pronunciation. You have to know Vietnamese then when I speak you can understand the standard pronunciation, but if I speak vocabulary, then in essence, if I speak English, it will be different from the Vietnamese I told my friends.


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

[espeak-ng:master] New Comment on Issue #954 Espeak takes too long to phonemize on colab
By Jaskaran170599:

I think version is the issue it does work fine on the 1.50 version but I need ubuntu 20.4 version for that and i want to run it on colab , I'll try installing the 1.50 version on colab by the installing instructions. these are the commands:

!sudo apt install espeak-ng #1.48.03 04.Mar.14 Data at: /usr/lib/x86_64-linux-gnu/espeak-data !pip install phonemizer

import phonemizer text=phonemizer.phonemize(txt, language=lang, backend='espeak', strip=True, preserve_punctuation=True, with_stress=False, njobs=1, punctuation_marks=';:,.!?¡¿—…"«»“”()', language_switch='remove-flags')


[espeak-ng:master] New Comment on Issue #954 Espeak takes too long to phonemize on colab
By Jaskaran170599:

I think version is the issue it does work fine on the 1.50 version but I need ubuntu 20.4 version for that and i want to run it on colab , I'll try installing the 1.50 version on colab by the installing instructions. these are the commands:

!sudo apt install espeak-ng #1.48.03 04.Mar.14 Data at: /usr/lib/x86_64-linux-gnu/espeak-data !pip install phonemizer

import phonemizer text=phonemizer.phonemize(txt, language=lang, backend='espeak', strip=True, preserve_punctuation=True, with_stress=False, njobs=1, punctuation_marks=';:,.!?¡¿—…"«»“”()', language_switch='remove-flags')


[espeak-ng:master] New Comment on Issue #954 Espeak takes too long to phonemize on colab
By Jaskaran170599:

I think version is the issue it does work fine on the 1.50 version but I need ubuntu 20.4 version for that and i want to run it on colab , I'll try installing the 1.50 version on colab by the installing instructions. these are the commands:

!sudo apt install espeak-ng # 1.48.03 04.Mar.14 Data at: /usr/lib/x86_64-linux-gnu/espeak-data !pip install phonemizer

import phonemizer text=phonemizer.phonemize(txt, language=lang, backend='espeak', strip=True, preserve_punctuation=True, with_stress=False, njobs=1, punctuation_marks=';:,.!?¡¿—…"«»“”()', language_switch='remove-flags')