Important Update from Team Irish!
We know you guys have been waiting patiently for our course, so I'd like to get you up to speed on everything that has been going on behind the scenes :)
(Note: This update has been directly copied from our course's page on the Incubator. I just wanted to make sure as many people as possible see this)
We are moving from our usual Saturday update once again to bring you up to speed on some important developments that have taken place behind the scenes.
As most of you know, the issue that is causing us to remain in Phase 1 is the TTS (or lack thereof). So it brings me great pleasure to tell you that a solution has been found! screams of delight
Let me go into a bit more detail :) Instead of using a traditional TTS like previous courses that automatically generate audio for individual sentences, our course will use recordings of a real human. This will mean two things: 1) The audio will be of very high quality and 2) Not all sentences in the course will have audio. But (<-And that's a big but) there will be more than enough sentences that have audio to make sure that every lesson has a good mix of different exercises. We are very happy with the audio we have heard and we would like to take this chance to thank Team Duolingo for their hard work and for getting us such a great voice :)
So, what does this mean for you, the 20,000 users that have shown their interest in our course? (See how I worked that number in real nicely without sounding too braggy?) Well, this means a lot of things! 1) We are well and truly on our way to beta. 2) Our course is going to have audio. 3) Our course is going to have really good audio. 4) You are gonna love our course and 5) You can get excited again xD
Estimated Time of Arrival: Soon™ (I know how much some of you hate that word, but let me assure you that now more than ever, do we actually mean soon...so like....really soon)
EDIT: All recordings are complete.
Beir bua agus beannacht,
AlexinIreland and the rest of Team Irish
Yes, yes, yes! This is fantastic news, Team Irish! Wow, the first course with a real human voice, amazing. Now I really cannot wait to try Irish out. :D
AAHH!! This image is driving me insane!!
Is it SO hard to draw a rainbow with the Red on the outside!?! is it!?! It is a simple fact of science!!
I hate it when ppl draw rainbows in the wrong color order or scheme or whatever.
So maybe it's a secondary rainbow then. ;-)
I know what a secondary rainbow is, but it is obviously not that.
If someone here is going to be a smart-alec, it is going to be me!
That's nothing. What about the people who draw it blue, red, green, purple, yellow, or some other combination. This rainbow is as close as we can get to the real thing. ;)
That's awesome...<h1>A real human :D</h1>
This course will have the best pronunciation yet...
At some-point it'd be interesting if the courses already offered (German, Spanish, Italian, etc) also implemented real human audio and we had an option between the text-to-speech or human audio.
I think everyone would choose human audio :) I don't think it's that necessary, like the Irish course, since they are understandable.
I think everyone would choose human audio :)
Careful, don't tell the Duolingo team that. When Duolingo was first being developed, there were human voices. However, Duolingo's "metrics" said that people preferred TTS over human pronunciation, and so that is what they went with.
I didn't know that! Where did you get that info? Any links/screenshots that I can check out? They have been using TTS since the public launch, so have you been here since their Beta phase?
I originally signed up during the public launch (this isn't my original account) and I never got to try the real voices, but here is what Luis von Ahn said.
I can't believe I found that source tbh. Had to CTRL+F "TTS" through all of Luis' comments. This information isn't widely known it seems, and I can see why. They rarely mentioned it.
Thank you very much, Zach! Luis should probably conduct the A/B test again since the website wasn't as great as it is now. I'll make sure to mention this to Monica in the incubator. :)
Really? What Z. Shan said, is there a source? I was just thinking that everyone would choose human audio, because there have been a lot of complaints about the voice, bad pronunciation, etc.
edit: Thanks for the link, wow, that was really interesting...
I, for one, am willing to wait any amount of time for this course, especially since the audio will now be from a real human! I'm jumping up and down with excitement right now! Thank you Alex and Team Irish for all your hard work! Luck Maith!
what region/dialect would he/she come from? :) And by the way thanks this is the most promising post yet haha :)
We do not know much about the person who is making the recordings, but we can confirm that she has quite a neutral accent that is very easy to understand.
So I take it that you have spoken with or heard her already, yes?
Moreover, I am guessing that she is a native of one of the Gaeltactht, yes? Otherwise a learned pronouncation may not be quite as 'good/ natural' (or maybe it would, I'm open to correction).
We have hear her voice in several samples and her pronunciation is pretty spot on. We do not know where she is from, but she does have quite a "neutral" accent.
Actually, come to think of it, the Professor of Irish in Cambridge is from the US, but I've been told that she has the most perfect accent as gaeilge (so much so that hers is a whole lot better than mine!), so perhpas it does not matter if the voice person is native.
I agree, being a native doesn't mean much. My French professor in college was not a native but when we went to France together, people gave her compliments on her accent. She said that because she has musical ears she can pick up accents easier and better than the average human.
WAIT, even though you can't get audio for every sentence, can we still at least hear audio for each word whenever you hover over it?
Like maybe you could cutout the soundbyte of the word from each sentence?
I don't have that info right now, but I will ask a member of staff. (They won't be back til Monday)
We don't know who they are, but we are considering giving them a name to identify them with :)
To throw out a few other suggestions:
- Gael (Gaelic-speaking person / inhabitant of Ireland)
- Ide/Ida (~thirst for knowledge)
- Deidre/Deirdre ("she who chatters" but also seemingly a rather depressing mythic character)
- Cady (~rhythmic flow of sounds)
- Teagan (little poet)
- Colleen ("girl" although it seems like expats created it as a name)
I don't suppose there's a cute name for "owl"?
WOW! I'm really excited for this course now, I was worried that the TTS would be as bad as French or as fuzzy as Italian, but it sounds like it might turn out for the best.
A REAL HUMAN
THIS IS GREAT NEWS! You have no idea how excited I am. Not only for the course itself (Brian Bóroimhe!) but finally... finally... there will be real, human voices on Duolingo! I hope that carries over to other courses eventually. I know what Duolingo's "metrics" have to say about it, but I think it's the vastly superior solution.
Thanks for the update! Soo, based on other comments, does this mean that the human doing the voices is not going to be from Team Irish? Is a professional then going to be paid for the voices? There was a confirmation on female, so is this a confirmation that somebody is already picked up, or just that the whole team wants a female voice? (Along with this is, is it even decided what method? Or because just "female human voice" with no other specifics?) I also think this is great news for other courses, since I know one of the Hungarian course makers was unsatisfied with Hungarian TTS. Thanks as always for your hard work Team Irish. I will be patiently waiting for the "soon" to come.
Thanks for the update Alex, this is amazing news! (Although I must admit I did pop into your chatroom and see this discussed recently so it's not really "news" for me). I can see this making the Irish course particularly popular and indeed quite special. I'd love to see this rolled out to every single course on Duo one day!
I have two questions though. Will all individual words have audio or not? What is the decision to select a sentence for audio?
Anyway I can't wait to try out the course, thank you again to you and your whole team for doing such amazing work :D
Thanks @_gs9! I thought I saw you popping in every now and then :P
Will all individual words have audio or not?
Each word will have at least 1 sentence with audio.
What is the decision to select a sentence for audio?
Team Duo decide that based on a variety of different factors.
Feel free to come by our chatroom anytime if you have any more questions, or even if you just want to say hi!
Haha! Thanks for the reply :) Maybe next time I'm having a nose I might just say hi!
Have my upvotes, have my lingots, have my undying adoration, and whatever else you want!
I do have a couple questions, though: Firstly, what dialect will the speaker use? My understanding is that Irish can actually vary quite a bit by location. Secondly, have you put any thought in as to what you'll do with the voice in the future? Do you know now if you will eventually get more recordings, potentially even with different voices?
I just screamed so loudly and hysterically that my windows are TOTALLY SHATTERED. Does anyone know any good window repairmen? Wow wow wow, the first course with a REAL HUMAN VOICE! I am so amazed!!! I AM SO EXCITED!!!!!!!!!!!!!!!!!!!!
Very excited for this! When you say not all sentences will have audio, I must admit I am struck with a certain sinking feeling...are there plans to eventually implement audio for all sentences, like before Irish graduates from Beta? The audio being the part I am looking forward to the most for Irish as its complicated phonetics are lost on my simple Anglophone mind ;)
Only time will tell. For now, we have a very good percentage that have sentences. (About 1/3). The issue is that our sentences will continue changing so it would be hard for Duolingo to always have audio for every one.
This is interesting. Personally I think it would be better to have both, the main "human" voices as the default, and the best possible TTS engine as the fallback. That way all sentences can have a voice.
Though that may cause consistency problems, but anyway, we're generally taught a language by people with many different accents and pronunciations, so I don't see it as a huge problem. But who knows what Duolingo's metrics will say about that.
P.S. Regarding the soon™, I think the incubator teams should come up with their own terminology or risk falling under the same trap as Duolingo. Rather than say Soon™ perhaps they should,in my opinion, say "within this year", or "next year at most".
We use "soon" because it actually will be "soon". We have an official estimate that is more accurate but we have been advised not to share that in case any issues arise.
I didn't say anything about the estimated release date lol, so not sure why you are responding to me. Anyway, I was talking about Duolingo's "metrics" which they like to reference so often. When Duolingo was first being developed, there were human voices but Duolingo's "metrics" said people preferred TTS over human voices, and so that is what they went with.
Edit: Here's the source of my information, for those who are interested.
I managed to escape my chains and came to roam around the main site for 5 minutes :P
You replied to my comment though.:P So that is why I was confused. Anyway, what are you doing here? You are supposed to be slaving away on the Irish course at all hours.:)
The general rule is that any estimate that is given must be multiplied by 2. So whatever estimate take it with a lump of salt.
Hofstadter's Law: It always takes longer than you expect, even when you take into account Hofstadter's Law. — Douglas Hofstadter, Gödel, Escher, Bach: An Eternal Golden Braid
This is why I never plan for some 'saftey cushion' time, because I know that I'll just end up exceeding that too!
Both would probably be way too expensive considering the estimates Luis gave for TTS, how TTS doesn't use a pay-per-sentence pricing model, and how expensive professional human translations cost.
That's a good point. Given a choice, I'd prefer to have a mediocre TTS engine for all sentences, and set up a special area in immersion where people can hear real "Irish voices" at their leisure, perhaps reading paragraphs or whole texts, and maybe even some including the words taught here.
That would be an inexpensive solution given that there is probably lots of public domain audio material out there, and lessons/concepts. Also, all these materials could be hosted off-site reducing server costs to Duolingo.
I just remembered it won't be any problem at all, people can always use their own browser to enable an Irish TTS voice, regardless of what the team chooses to implement. I'm just not sure if there are such options for mobile devices.
Purely down to financial constraints. Voice artists charge per sentence recorded, while no such limit applies with a TTS system.
I can check with our mentor who has all this info and I will let you know on Monday when they return.
Not necessarily :) I'm not confirming that it will or won't, our mentor is just on a break. If the Duo team are ready, then we'll launch with the help of another mentor who is helping us atm.
I'm sure there are logistic considerations to think of. It takes a long time to get the recordings done, reviewed, edited and then embedded in the software. If they were to do every sentence, we would be a very long time waiting for the course.
Thank you for the update alexinireland,
Can you recommend any sources to learn the Irish alphabet? I would like to become familiar with each letter and their sounds before the course is released.
Just make sure you learn all the mutations and how the lenition and eclipse system works. Also, slender and broad.
I am so ready! Well, not really, because I don't know how to handle three langs, but I don't see that stopping me or anyone else here, right?
Are they a native speaker? If they're not, do they have the palatalized/velarized (broad/slender) distinction of consonants down?
We do not know if they are native or not, but their pronunciation is spot on.
I am very glad - this will probably also put the standard for other minor languages which also will have the same problems concerning the sound.
> our course will use recordings of a real human
I hope this sets a new standard.
This sounds amazing, I can't wait. Just out of curiosity, how good of TTS is www.abair.ie? Would that have been useful as a TTS engine? I'm sure you've already explored this option, but I just wanted to see if you guys think it's going to be a useful learning tool going forward.
Thanks again for all your hard work! :)
It seems like Abair was their first option, but they had some problems with implementing it, so they were not able to use it for their course.
Well, isn't this great? It makes me wonder what other courses will have a real human voice… * fistbump *
Recording all the sentences will probably take some time, so no, I won't get excited again until I see that the course is released. But thanks for posting updates every few days, it makes it so much easier for me to wait when I know what's going on behind the scenes. :)
Just note that quite a few of the sentences that will have audio have already been recorded.
This is so awesome! I have been waiting so long for this course! A real human voice will sound a lot more crisp and clear. Thanks so much!
I, of course, am reading this right before I am about to go to bed. Which was a horrible idea because I AM SO EXCITED!!!
You and me both. It doesn't help that I'm already about five hours late for bed!
I just wanted to finish that French skill off so badly before calling it for the night, and then I simply had to check the news in case there were any updates on Irish once I finished it... Oh well!
Glad to hear it guys and girl. Good luck with the final touches. I'm looking forward to the course!
This is all great and fantastic news, I can't wait! But the best thing of is post, is the fact that you use Soon™ as your trade mark, hahaha! :D
Team Irish, that's great news! But, what I'm more excited about is the fact that Irish's spot in the incubator is likely going to be replaced by a new course!
This is what's been happening lately. Look at Ukrainian for English, as soon as it got out of the incubator, Italian for French popped right in. Come onnnnnnnnnnn Hebrew!!
Plus, you'll have the Irish course to work on while you wait for the next one ;)
Screams of delight is putting it mildly! It's like the holidays, seeing these updates and knowing the Irish course is close at hand. So. Happy.
The Irish section of Forvo is very good. The are real people recording, not TTS. Only standalone words allowed though, not sentences.
I know of a few Irish contributors including yourself on Forvo; are there any you'd recommend specifically for Ulster pronunciation? I don't recall having come across one, and it'd certainly prove useful for me to familiarise with it.
I can only think of two, Lughaidh and AnTreasach. Unfortunately neither are contributing any new material.
Which dialect are you going to use, or will it be all four? I have a basic understanding of Irish, and have a cúpla focail, but Ulster Irish sounds like a completely different language to me
Don't worry, Ulster Irish sounds like a completely different language to most Irish people too!
Well, maybe not that bad, but it is quite different, and takes some getting used to. I don't think I've ever heard anyone speaking it in real life though, I've only ever heard it on the Irish tapes, and maybe the odd time on the radio or tv.
I have, my Dad lives in Tyrone and his uilleann pipes tutor is a gaeilgeoir, I can barely understand what he says. I'm often saying to him "Stop a labhairt uladh ngaeilge, ní féidir liom a thuiscint leat!" That about sums up the extent of my Irish speaking abilities haha
Abair is a good resource, so don't worry if you are using it! There were difficulties implementing it with Duolingo's system, so we went another direction.
If your team will have really great audio, then shouldn't the other courses use whatever software you're using?
They aren't using TTS software, a woman is recording the sentences. Personally, I'd love for human pronunciation to be a part of every single course.
Actually I would think that would be best of course, but didnt want to suggest it because I thought it might be too much of a hassle. Now that I reread it though I see they are doing that, is awesome!
Only the lack of voice in the majority of the questions makes it very difficult to retain, unlike the other courses where the word or phrase is always spoken. As result I have a nasty habit of mispronouncing the words. I very much prefer it to speak every time.