Duolingo is the most popular way to learn languages in the world. Best of all, it's 100% free!

https://www.duolingo.com/breravin

Swedish audio: TTS software vs. native recorded

I know this is a big task/ask, but I think one of the first "stretch goals" for the Swedish team should be to record actual, spoken Swedish. The TTS (computerized text to speech) is just too lacking, especially when genuinely trying to learn a language and become familiar with the pronunciation.

I (and I assume many) would actually be more than happy to pay for such an addition. I know it doesn't work that way (i.e. pay for features), but my point is that there may be ways to offset resource limitations. For example, I have connections in the sound/audio industry who may be able to recommend talent and actually even arrange recording sessions.

So far I'm pleasantly surprised by how good the Duolingo format is. Please consider adding recorded audio and increase that quality even further!

Update: there are some very good points about the benefit of TTS over live recorded below. I'll keep the original post, but I get why TTS is a good way to go. So hopefully instead we'll see some improvements to the TTS (in the long term)!

3 years ago

22 Comments


https://www.duolingo.com/Carnaedy
CarnaedyPlus
  • 19
  • 17
  • 12
  • 10
  • 6

Audio for the Irish course is a live recording, and it appears to have a lot of technical issues attached, for example, not all words have their pronunciation separately, there is no option to rewind the sentence slower, etc. While I agree that hearing a person speak the language would be awesome, it is not as trivial as we might think, if we want this done properly.

3 years ago

https://www.duolingo.com/breravin

Ah, good points - thanks for the perspective :)

3 years ago

https://www.duolingo.com/superdaisy

Also, words recorded individually are going to sound way different from words recorded in a sentence. At least in English, we tend to have a spillover of sounds that can't be smoothly clipped or strung together.

3 years ago

https://www.duolingo.com/Anders91
Anders91
Mod
  • 19
  • 15
  • 12
  • 11
  • 8
  • 8
  • 7
  • 7
  • 6
  • 5
  • 3
  • 2

Using live recordings is not only very time consuming; it also comes with a lot of disadvantages.

Using live recordings you are not able to listen to the words slowly, something I have found very helpful when using Duolingo to learn Portuguese, also, you cannot listen to words separately.

We are looking for better TTS software for Swedish and hope that this will fix any problems we have with the audio at the moment.

Right now, I do not see pre-recorded audio being used for the Swedish course.

3 years ago

https://www.duolingo.com/jbailey88
jbailey88
  • 25
  • 16
  • 13
  • 12
  • 2

Could you use the Ivona one - http://www.ivona.com/en/ ? Although I'm not a Swedish speaker, the TTS does sound very natural. Plus, all of the voices for the languages that I do speak seem to be very high quality.

3 years ago

https://www.duolingo.com/feanarosurion

That sounds fantastic, I think it would be a big improvement over the current voice! I notice right away it fixes the issue with "de is really pronounced dom" that the moderators posted about. Hope they could make this one work!

3 years ago

https://www.duolingo.com/Adelost
Adelost
  • 10
  • 8
  • 6
  • 4
  • 3
  • 2
  • 2
  • 2
  • 2

I concur, Ivona is the most genuine sounding speech engine I've heard so far.

3 years ago

https://www.duolingo.com/breravin

Wow, that really sounds good. By far the best TTS Swedish I've heard. Thanks for sharing :)

3 years ago

https://www.duolingo.com/Anders91
Anders91
Mod
  • 19
  • 15
  • 12
  • 11
  • 8
  • 8
  • 7
  • 7
  • 6
  • 5
  • 3
  • 2

Thank you for your suggestion, we are considering every option out there.

Ivona sounds a lot like the Swedish queen in my opinion heh.

3 years ago

https://www.duolingo.com/HelenCarlsson
HelenCarlsson
  • 23
  • 18
  • 18
  • 18
  • 15
  • 15
  • 12

Mina barnen är så trevligt :)?

3 years ago

https://www.duolingo.com/jake3389
jake3389
  • 18
  • 15
  • 14
  • 13
  • 13
  • 12
  • 11
  • 10
  • 6
  • 2

Ivona is probably the most voice realistic TTS out there, however I would prefer the TTS from Acapela (http://www.acapela-group.com/?lang=sv) which includes TTS in four dialects and both male and female speakers.

3 years ago

https://www.duolingo.com/sualainnis
sualainnis
  • 23
  • 11
  • 7
  • 6
  • 3
  • 3

Is there any point in reporting incorrect audio (is it something you can fix)?

3 years ago

https://www.duolingo.com/breravin

Once the limitations of live recording were mentioned, it made a lot more sense. I was under the impression that it was more an issue of limited resources for actually doing the recording.

I find it super useful to listen to native speakers, but Duolingo is perhaps not the right place for it. Certain languages have come really far with TTS, so here's hoping for some more natural Swedish TTS in the future!

Thanks for the follow up!

3 years ago

https://www.duolingo.com/HelenCarlsson
HelenCarlsson
  • 23
  • 18
  • 18
  • 18
  • 15
  • 15
  • 12

By the way, is it possible to modify the pronounciation of certain words (without changing to another voice)? If so, how does it work (I am just curious)? For example, I just found out that "gaffel" (fork) is pronounced as "gafell" which of course is very confusing.

3 years ago

https://www.duolingo.com/Anders91
Anders91
Mod
  • 19
  • 15
  • 12
  • 11
  • 8
  • 8
  • 7
  • 7
  • 6
  • 5
  • 3
  • 2

To be honest, I don't know. All I know is there are no tools for the TTS available to us contributors in the incubator.

3 years ago

https://www.duolingo.com/sntindall

Yes I think just a better TTS is what's needed. It would be great to hear a native speaker but I feel like maybe Duolingo isn't the best place for it. I have found the slowing down of the TTS has been very very helpful, it would be even better if I could understand the TTS more.

3 years ago

https://www.duolingo.com/G0108

Why not at least make fix it (if possible), the pronunciation of words like äter, är, and de is not natural to me.

3 years ago

https://www.duolingo.com/Its-Me-Me-Me
Its-Me-Me-Me
  • 20
  • 10
  • 7
  • 6
  • 4
  • 3

I find the current Swedish audio quite difficult to understand and follow but I'm sure I'll get used to it eventually!

3 years ago

https://www.duolingo.com/HelenCarlsson
HelenCarlsson
  • 23
  • 18
  • 18
  • 18
  • 15
  • 15
  • 12

It often helps a lot to listen to the slow version. I do that all the time with Danish (and realize that the sentence has at least three more words than I thought :)).

3 years ago

https://www.duolingo.com/sntindall

yes this too! The tts pushes like 3 words together sometimes! Frustrating!

3 years ago

https://www.duolingo.com/HelenCarlsson
HelenCarlsson
  • 23
  • 18
  • 18
  • 18
  • 15
  • 15
  • 12

Danish is worse, believe me :)! And there, you can't even blame tts.

3 years ago

https://www.duolingo.com/sntindall

I had a hard time with it to start with but the more you do it, the easier it becomes. Even if you still have a hard time understanding you'll have a basic knowledge of what the word might be so that helps. I've actually been opening google translate in a separate tab and putting in what I think the sentence is to see if I'm close so I don't lose a heart! It's helped me learn more quickly.

3 years ago