Hello everyone,
and welcome back to your favorite German learning website.
So many of you probably remember that I partnered up with EF languages for their speech recognition system, and that I am working in integrating that more and more here.
I think speaking actually gets WAAAAYYY too little attention in all the various online courses and systems. I mean, ten years ago, the tech wasn’t really evolved enough, but we’re in 2021 now and there’s really no reason that “fill in the gap”, “multiple choice” or “order word boxes” should be the backbone of exercises these days. I mean, they do have their place, no doubt, but there’s a lot of potential in making people actually speak.
I created two little workouts with the EF exercise toolkit earlier this summer, and you seemed to have liked it as an idea. But there were a lot of small functionality and design thing that were missing or that I wanted to change, to make it a REALLY nice quiz experience.
And so I went all Thanos, like “Fine, I’ll do it myself.” and I actually started programming and over the last few weeks I integrated the speech recognition into the normal quiz setup that I am using.
It’s not 100% finished yet and there are plenty of features I want to add, but it actually works, and so I really really wanted to show it to you and get your feedback :).
So today, I proudly present…
My New Quiz Setup – With Speaking
If you’re a regular reader, it’ll look pretty familiar to the other quizzes on the site. It’s just that now, some questions have to be answered by speaking.
I’ve setup a short quiz for a start about using “ankommen auf” in the sense of “to depend”. It’s just seven questions, but it gives you a good impression of how it works and what you’d like changed or improved.
Here’s how it works:
To record, just press record and then stop. The AI then checks how well your recording matches.
If you’re below the threshold (right now at 70%, but eventually you can set that yourself), then you’ll be asked to try again. If you’re unsure about what you’re saying is even correct, you can check the hint or listen to my version as a reference.
If your version is better than the threshold, you’ll see the proper solution and a word by word breakdown on how well your answer matches each of the word.
“red” and yellow“ mean a not so good match. Either your pronunciation was really bad or you just said a wrong word or forgot a word.
“light green“ means that it was decent, but needs work and“green“ means that it was pretty darn good. And “green with frame“ means that it was 100% perfect native level.
NOTE that I don’t always score 100% either. The mic quality plays a role, as does the surrounding noise and of course the AI isn’t perfect either.
Anyway, if you get a pass, you can either move on to the next question, but you can also try as many times as you like to maybe get all the words to green level. Just hit record again.
And of course you can listen to yourself and compare all the recordings and their grading.
Oh and the recordings are kept till the end of the quiz, so you can review all questions AND your recordings after you’re done.
Pretty cool, right :)?
In the future, I’ll also add the option for you to set the percentages for the colors, so if you want a really strict AI that only lets you pass above 90%, you’ll be able to do that. But for now, I think I’ve set it to a good middle ground.
I also want to add the option for you to switch to typing, and I want to give you an average of the pronunciation quality on the result screen, but hey… the main focus of the exercise is that you practice German words and structures, not that you get a poster perfect pronunciation. You do need native speaker feedback for that, after all.
Anyway, so yeah… I think we’re ready to jump in, so … viel Spaß mit der Übung :)
And… how did you like it :) :) :)?
Please let me know in the comments how you liked it and if you liked this better than the first two exercise options from a few weeks ago.
I’m sure there are glitches, too, so please let me know ALL the feedback. What you liked, what didn’t work and what features you’d like to have. And be wild there. Because it’s now quite integrated into the site, I have a LOT of control over what features I can add and how to add them.
So yeah… I really can’t wait to read all your thoughts and comments.
Thank you all for reading, have a great week and I’ll see you soon… (when I announce my vacation ;))
further reading:
Practice Speaking – Past Tense (similar idea, but less customized)
Practice Speaking – Three verbs (second version, different, but not yet ideal)
This is way cool!
Glad you like it :). There are already a few more exercises like this out on the blog. I have to make an archive for them so they’re all in one place.
This is great! And thanks for the member sponsorship!
I think the scoring at the end is a too lenient because I got 115 out of 116 points and yet made a lot of mistakes.
I’ve used a similar tool for speech testing in an Italian course that I took (Wellesley intermediate Italian via EDX, free if you want to try it). Each time you spoke it produced three or four sentences showing how the computer interpreted what you said. That gave me a way to focus on how the sounds I was producing were off. By contrast the color scheme in your tool lets me know which words are problematic but not why they’re problematic. Plus it also gives away the answer. I don’t mean to be critical in the negative way, it’s just that the more specific feedback given, the better.
Is there anyway to install little screens on every Germans forehead that shows in real time what they’re hearing? That would be the most helpful.
The scoring is 100% all the time, unless you skip a question. I have plugged this AI module into my normal quiz setup and the two don’t really work well together.
Maybe I should hide the points all together.
As for the showing the answer… you can fix that by adjusting the difficulty upward. If you set it to 90% you’ll likely fail more often and not see the answer.
The system I am using is bound to one sentence, so it cannot freely transcribe what you say.
I can see how that is useful, but there are limits to it, as well. A native speaker’s brain will fill in a lot of gaps from context whereas a computer AI that transcibes speech can write down some really weird things.
At the end of the day, if you want to get qualified, detailed feedback on your pronunciation you need a native speaker.
The point of this exercise is not practicing pronunciation, it’s practicing to say stuff in the past tense.
I think that speaking is not used enough for exercises online, except for pronunciation. And that’s absolutely NOT what it’s for, in my opinion :).
Nothing beats a native speaker’s ear there.
This was really cool! Thanks for working so hard on it! :)
Thanks you. New one coming in a few days, about past tense :)
Cool quiz. I also had trouble with a few words, like “essen” strangely.
This is really cool! I have issues with it recognizing that I am saying certain words. Not sure if its my pronunciation is that bad or its just the AI.
It does have problems with a few words for me as well … so it would score me at like 70% for a word consistently, which it shouldn’t as I am a native speaker.
How low did you score on these and which words were they?
Hat irgendwer mit einem Mac diese Übung zum Laufen gelingen? Ich habe mit Safari, Brave und Chrome probiert. Ich kann auf dem Unten den Voice-Recorder mit Erfolg verwenden, aber in der Übung kann ich nicht es aufnehmen. Beim Untersuchen habe ich verschiedene Probleme mit dem Audio-Recorder-Modul gefunden.
audio-recorder.min.js:1 Uncaught TypeError: window.onresize is not a function
at HTMLDocument. (audio-recorder.min.js:1)
(anonymous) @ audio-recorder.min.js:1
audio-recorder-AI.min.js:1 Uncaught TypeError: Cannot read property ‘classList’ of undefined
at y (audio-recorder-AI.min.js:1)
at HTMLDocument. (audio-recorder-AI.min.js:1)
y @ audio-recorder-AI.min.js:1
(anonymous) @ audio-recorder-AI.min.js:1
Emanuel, wenn du mit meinem Mac Laptop testen möchtest, lass mich wissen. Ich würde gern helfen.
Kleine Korrektur: “zum Laufen GEKRIEGT” nicht “gelingen”.
Thanks a lot for the bug report!!
I added some functions yesterday (something about submit-buttons) and I actually made a logical mistake with counting through an array.
I can happily say it is fixed now :)! Maybe you need to do a full reload though, as I am caching most of my js.
Ich werde es heute Abend checken und zurück berichten.
This is awesome. Super!
I want to add that my voice was not recorded using Safari so I had to switch to FireFox to get that part to work. If someone knows what to do to Safari please post a solution since I prefer using Safari as my browser.
Did the buttons work, so did you get to record silence or did the recording not even start?
I really liked the implementation. It worked very well for someone like me who does not ever speak German.
Cool, glad to hear that :)
This is an incredible feature! I’m amazed it works so well too
Glad to hear that :)
I use a Mac. The hint function works and so does the multiple choice responses but the recording button doesn’t respond. I can’t seem to record anything. Is this a fixable glitch?
Hard to say :/. Do you know how to open developer tools on the browser? I don’t have a Mac and it’s virtually impossible for me to fix this remotely, as it is device and browser specific.
The one thing you could do is try a hard refresh of the page.
You can find how to do it for Mac for different browsers here:
https://fabricdigital.co.nz/blog/how-to-hard-refresh-your-browser-and-clear-cache
Viel Spaß. Danke. That was cool.
Meine Schwäche ist den Unterschied zwischen bitte / bitter. Grrr…
Wettahh.
DAS kommt darauf an, …
aber in dem letzten Satz: ES kommt darauf an, ob du Fisch magst.
Warum DAS —> ES ?
I kept switching in the examples between “It depends” and “That depends”. It’s a bit confusing for a learner because you have to pay close attention when you actually should “flow”. But yeah, some examples are with “that/das” others are with “es/it”.
DAMN ITS SO ACCURATE, i wasnt expecting that
Working great with Chrome on Windows with my soundcard and mic.
Much, much better! Couple of minor niggles / suggestions… (1) when you get it right it seems that you have to hit next twice to advance to the next question (2) Did I miss the button to hear the correct sentence? (3) On one of them, I was having trouble coming up with the right words, but there didn’t seem to be a way to give up and just hear the answer. I eventually got it, but seems like I could have got stuck?
1) Yes, that’s because I hacked all this into my quiz software (third party) and the “submit” button has to be there so you can see your result for the multiple choice questions. I can hide the button and auto-push it, but I need some time :)
2) It’s should be visible when you open the hint
3) Again, did you check the hint?
You aced this! I struggled with the early version but found this was super easy to use. Thank you so much for spending the time to perfect this. It’s going to help me a lot.
Sweet, glad to hear that, thank you :)!
As usual, (and as per the other “practive speaking” modules) I could not get this to work at all. I suspect that I can’t get my mic to connect or something, because it caught virtually nothing of what I said. Sigh. Can I just slide down into a younger generation so that this isn’t always such a struggle…?
I’m sorry to hear that. Maybe I should do a little tutorial video at some point, but the issue is that everyone has different devices, browsers and operating systems. What works for my PC won’t work for Mac and so on… anyway, sorry for the frustration!
It’s ok, I’m used to it. I really enjoy the recoder attached to your lessons and use it a lot. That works for some reason… go figure.
Wait, that works? That’s really really weird because I am using the same code for both pretty much.
What exactly happens for the quiz-recorder, when you click?
5 out of 7 said incorrect and then 2 seconds later it said awesome that’s close enough. I’m in Grand Rapids Michigan so maybe I’m in some kind of time delay Anyway I enjoyed the exercise It’s been 40 years since I taught German
Just to make sure… you do NOt have to press submit to send your recording. The recording sends automatically when you click the recording button again.
I think I’ll remove the “submit” button, because it’s confusing. It’s just there for the multiple choice questions, but I’ll find a hack around :)
Oh, so you can use the recording feature all over the site now, neat! I gave it a whirl and it works well. Just a small heads up, there’s a typo when you get over the 70% here, it says “aweseome.”