Audio from human speakers
As I am using Duolingo I can't help but think that the digital voice that is speaking during the examples, probably isn't how native speakers speak. So I thought, 'wouldn't it be nice if each example came with the current audio, as well as another track spoken by a real person?' In order to do this Duolingo could ask for submissions from its community. If they really wanted to go the extra mile, they could provide even more tracks by native speakers that would provide examples different regional dialects and accents. An example would be like Bosotnian - English and Southwestern US - English or Kanto - Japanese and Kansai - Japanese.
This is the only thing I want right now. I feel that it could really take these courses to a whole new level.
I've actually had an idea for a long time on how to go about doing this. Users submit audio recordings for text that Duolingo requests and they have someone sort through them to sift out any that don't adhere to the sentence. Or use the sound detecting software they use currently. I don't know.
These selected recordings are then moved to a section on the website where users can attempt to translate them based on what they hear. Sort of like Immersion, but with sound. The ones that get the highest amount of correct translations are then used for lessons.
It might work better if the user submits their sentence of audio with a caption stating what the sentence says. Then, similar to immersion or discussions, users vote up the sound of the audio, accuracy of the pronunciation/grammar/words/, and obviously, as you said, accuracy of the translation. So, very similar to what you said except the user submits the caption with the audio file, and without the filtering system?
I think the current voice is much clearer that a native would be. It may not be that authentic, but it's better to understand, and it isn't that bad in my opinion (the German at least, can't say for others). Would you want to get marked wrong for an audiotranscription you never really can understand clearly? If you want a native spaker, you would need to get a professional speaker to do it, otherwise you'd probably learn some dialect or slurred speech.
But do you really think they would be "more natural"? Honestly, if I were recording a sentence for duolingo I would pronounce it WAY more careful than if I used the same sentence in conversation. Because in conversation, I talk to another native speaker. Who has context. And who can ask me to repeat or clarify if I mumble too much. I'd swallow syllables, I'd cange to course of what I want to say mid-sentence or repeat or change my wordings according to the puzzled look on the other person's face. Conversation is not only pronouncing a sentence it is watching your partners face and reaction, factoring in his age, the situation and sourroundings and so on. And when all other factors of conversation are taken away like in an exercise sentence, you will in turn start to over-emphasize ponounciation.
So tha baseline is, if you want to listen to how natives speak, watch tv series or travel to the country and listen to the people in the streets. A mumbled random sentence would not teach you very much.
Yes, they would be more natural. I don't know how good the spanish or german voices are, but I tried the french one (I'm French), and it's far from perfect. And by that I mean it is often wrong. If I had to submit recordings, I would probably over-articulate too, and it wouldn't be perfect, but I wouldn't make those mistakes. I could make mistakes, but they would be more natural too.
Moreover, learning from the artificial voice will not help you understand accents. Having various accents would help.
I agree, the German voice is good as it is. The Italian one still needs a lot of work but it too has potential. In the meantime, I supplement with Forvo if I question the machine's pronunciation of a word. Rather than spending time sifting through multiple versions of the same sentence, it would be preferable to me if Duolingo improved the spoken sentences they have now.