SuttaCentral

SC-Voice: feedback for fine-tuning Portuguese TTS voice (aka Ricardo)

sc-voice
Tags: #<Tag:0x00007f788ccc9f28>

#41

Yep :blush:
Google Translate did most of heavy lifting, I just revised it carefully. It is amazing how well it works for this kind of descriptive texts.
Note that I decided to translate the explanation of all the English voices available as I assume some of the Portuguese-speaking visitors of the website may be interested in knowing more about it.
:anjal:


#42

Thanks for letting us know! You seem to be comfortable with this kind of work! :grin:

Those fixes may not be in the new update yet. I am not sure, but I think Karl first has to push it to the staging server, and only then can I update. But the basics are all on production now!

For German it usually does pretty awful! :scream_cat:

This is how you see fit. Thank you so much for dog it!


#43

Great. So does this mean I can keep adding these the way I have been adding and eventually these changes will find their way through the official SuttaCentral Voice? (@karl_lew ?)


#44

Exactly, that’s how it goes!

Gabriel, I have done the Home page, the rest will follow as I find the time. Did you take the necesary screen shots? That would be helpful for me; for otherwise I have to do this. :laughing:

Thanks!


#45

No, I didnt. I think it is okay to use the original screenshots. People will make sense of it!


#46

If you want to edit anything in the Wiki you can do so anytime by pressing the “Edit” button at the top right. You have all powers over Voices GitHub, including the Wiki.

If you want to add the new pages yourself you are also welcome to do it; otherwise I will do bit by bit.


#47

Please feel free to do it and at the pace you are okay with. I will not get involved with editing the Wiki at this stage.


#48

Hi @karl_lew, I would like to ask your help identifying what is odd with the text below:

“‘Bhikkhu, uma pessoa consiste de dezoito tipos de examinação mental.’

It is found in MN140 and SC-Voice is struggling to read it and I believe it has to do with the odd combination of quote marks preceding the word bhikkhu.

If that is the case, what do you think should be the right text format. Just to remove the (double quote). Should we get @Aminah’s help cleaning the legacy Portuguese texts of these?

Or maybe I should try adding a customword to capture it?

:anjal:


#49

Sometimes this sort of combination of quote marks occurs in the texts, in cases where a person speaking is quoting someone else or so. This is part of the texts and can’t be changed.

If this is the cause for trouble with pronunciation, I think changes have to be made for Voice, not for the original texts. I remember that Karl had to do some adaptations in cases where Amy would read out parentheses or so, and I don’t know how, but he got it done.


#50

Indeed. I am aware as well that the quotes encoded when typing using a Portuguese keyboard/key map may not be necessarily the same as when using an English keyboard.

Maybe it is the case we need to replace what was used when I encoded the legacy Portuguese translations with something more standard and friendly to the TTS.

:anjal:


#51

In short; what Ang. Sabbamitta said! :smiley:

The quotes are not causing the problem. For bonus points, though, I will mention that it is a typographical standard on SC to use the correct punctuation for a given language. Checking what quotation marks Portuguese uses on Wikipedia I’ve discovered that the quote mark type doesn’t look to be specified for Portuguese (that is either round or angular brackets can be used). It further notes:

The Código de Redação [70] for Portuguese-language documents published in the European Union prescribes three levels of quotation marks representation («…“…‘…’…”…»):

    E estava escrito «Alguém perguntou “Quem foi que gritou ‘Meu Deus!’?”.» na folha de papel.

    And it was written “Someone asked ‘Who shouted ‘My God’!’?”. in the sheet of paper.

All in all, the quote marks in this file look to be correct. :smile:


#52

Good! Thanks for confirming! :blush:
Hope @karl_lew has a way to fix things on the TTS side of things (i.e. get Ricardo to read it right!).
:anjal:


#53

I was just about to close this file, but glancing at it again I figured it might be helpful to check if this problem only occurs on the line given above:

“‘Bhikkhu, uma pessoa consiste de dezoito tipos de examinação mental.’ (ln32)

Or if it also occurs on the two preceeding instances of a similar line as well:

“‘Bhikkhu, uma pessoa consiste de seis elementos.’ (ln30)

“‘Bhikkhu, uma pessoa consiste de seis bases de contato.’ (ln31)


#54

Yes, it is occurring whenever the TTS has to read aloud a sentenced starting with that combination of quote marks. And this happens all over the place.

I have just noticed that SC-Voice is also having issues reading the text in between square brackets in MN1:

… ele concebe [a si mesmo ] em …

It is reading the end square bracket aloud as vírgula, which means comma in Portuguese!


#55

Gabriel, please give me an exact reference (sutta number + paragraph number, like MN23:2); then I’ll make an issue for Karl to address at his convenience.


#56

mn140:1.0.19

For fun I tested out how ChromeVox handled this (a while ago, to my great amusement, I discovered that my English reader can reader all languages; with a comically English accent). ChromeVox does read “Bhikkhu” where it should, where Voice does not (I can confirm Gabriel’s report).

BTW, this is the first time, I’ve used Voice in another modern language; it’s simply brilliant!


#57

#58

Sure.

An example of the first issue is found at https://suttacentral.net/mn140/pt/beisert#wp8 and many of the following paragraphs.

And example of the second issue flagged above is found at https://suttacentral.net/mn1/pt/beisert#wp3 and many of the following paragraphs.

Thanks for your attention and help! :anjal:


#59

Indeed it is! Mindblown!


#60

Most excellent! Glad to see you are comfortable with IPA. I would very grateful if you would have some time to assist us with the other translators in their respective languages as their own translations come online in Voice. Others may not have your facility with IPA and would greatly benefit from your teaching them how to “think and hear in IPA”. I have only recently learned it myself and found it a wonderful exercise in body awareness, volition and mindfulness. :pray:

Yes. We can fix those things. These problems you found are not trivial to fix, so we’ll need to prioritize. Thank you for detailing the issues–we track all these in Github and you should feel free to add individual issues as they arise. All of the Voice admins have experience adding issues and we can answer your questions about Github and Zenhub. Mostly we need issues to be atomic and descriptively self-sufficient so that we don’t need to ask you anything other than to test the fix. Also note that since you are now part of the development team for Voice, we might assign issues to you for testing. :wink:

Regarding priorities and expectations, our primary goal for the next official release is to integrate with Bilara so that we can host Anagarika’s translated segmented Deutsch suttas as well as all the upcoming translations from your team as well. Thanks to Blake and Bhante Sujato, Bilara data format is simple and stable, so we will be able to do this work in advance of the Bilara UI. Anagarika Sabbamitta is currently writing text files that we’ll transform into Bilara-compatible data. If your team requires similar assistance in converting non-Bilara files into Bilara format, please do let us know.