Suggestion: Better Pronunciations Idea
So we all know and hate the text-to-speech currently implemented, but waiting around for native speakers to do something isn't a realistic or efficient way of handling it in my opinion. All we have are the "repeat after me" exercises, which give absolutely no feedback. However, once the video-subtitles and grammar suggestions are implemented, speech lessons/feedback will be the only aspect that duolingo will lack.
So why not let users provide pronunciations to phrases in the same way you handle crowd-driven text translations? Users would then be able to get feedback on their speech through votes and comments of higher-level members. Even better, the best pronunciations could then replace the speech synthesis, thus offering a more immersive variety of dialects and accents for each phrase.
This could be an addition to the article translation system, or an improvement to the lesson questions.
So you want non-native speakers to provide the audio? Sorry, but that's a terrible idea.
Agreed. And just for the record: I don't "hate" the TTS. I actually consider it to be pretty good, except for some occasional hick-ups. In my opinion the only way to improve the audio would be to hire professional native speakers. [And yes, duo should take that route eventually]
Speak for yourself. I know the text-to-speech system and I think it's fine. As I say every time one of these crowdsource-the-audio suggestions comes up: the TTS is clearer than any of the native speakers I encounter in either of the languages I'm learning. Occasionally there's a glitch in a specific sentence and I user the "report a problem" button to report it.
Crowdsourcing the assessment of people's pronunciation as you also suggest seems, at first, a more attractive idea. I'm not overly impressed with Duo's speech comprehension in German: usually I have to give up on a natural pronunciation and switch to a "robot voice". So yes, if the technology can't be improved, having a native/expert human assess my pronunciation would be great. But I doubt if there are enough native/expert humans to go around, and obviously it would require big changes in the system: you can't expect instant feedback if the assessment is going to be crowdsourced! It would be more like "pronounce this sentence, we'll get back to you and let you know if it was OK". Finding a Skype chat partner would probably be a better way of using humans to improve your pronunciation.
Actually, I'd find this very interesting. I'm an English as Second Langague teacher myself (although, not the best one, I confess), and I believe that understanding something is much more important than being able to pronounce something as a native. I find it very embarrassing to turn off my microfone because the TTS system doesn't recognize what I say 90% of the time while trying to speak French. I switched microphones, changed the way I pronouce the sentence and every time it wouldn't recognize what I've just said. Now, it may be a problem with my pronunciation, or not, but since I don't have any partner to help me, I'll never now. NOW, imagine if Duolingo could store our recordings (if we'd authorize it to) and send it to other people so they could type what they've understood. The system could work like this: In addition to the regular TTS exercise, one could ask to send his/her recording to another learner and see if it can be understandable. How would it work? The other learner would type what the person just said and, if the answer is correct, the audio would get +1 point. If the answer would be wrong, the system could analyze what went wrong: checking the written sentence if it'd have grammar mistakes, or, ask the second learner if the sentence isn't understandable. If the learner selects that the sentence isn't understandable the sentence would receive a (-1) point and the learner would receive a chance to say "why" the sentence is bad. This comment would be moderated, and sent to the original speaker as feedback (if approved). In that way, the sentences with more points would be considered clearer and more comprehensible, therefore, used more often and the worst sentences would receive (positive) feedback from several learners. To maintain quality, sentences could be discarted after reaching a minimum score, for instance, -5 and, the best sentences would be either replaced or approved by a native to keep as template for that exercise.