and welcome back to your favorite German website … I mean, hopefully.
And today, it’s again time for a new episode of
My New Feature
And this week, my new feature is… drumroll… my 2023 beach body.
Yes, it’s only March, but the Michelangeloian masterpiece is already finished.
Here it is:
That sweat in the gym was worth it. All three drops of it.
I’m kidding of course.
This is not my belly at all. I am actually incredibly shredded.
By new feature I mean of course a new feature for the site.
You might have noticed (or will notice soon) that the buttons for the examples have moved and it looks a bit more clean now.
But that’s not all. The real update was for the AI pronunciation trainer.
The New Pronunciation Trainer
So many of you (I hope) are probably at least somewhat aware of my pronunciation trainer.
It’s using a speech recognition AI by EF Languages (shout out to the engineers there) that grades how well you pronounce something based on rhythm, melody and the quality of the sounds themselves.
I have a few dedicated exercises for it (the series on how to use the most important verbs) but I also had this tool where you could load pretty much any example from the site and practice it.
But the tool would open up in a separate little box, taking up screen space, you couldn’t fully see the longer examples, and I got several calls from Window 98 who wanted its design back. I was never really happy with it.
So now, have finally upgraded that and it’s MUCH better now.
Because now, the trainer opens right under every single example.
Here’s essentially how it looks now:
- Thomas hat sich im Meeting die Hose ausgezogen.
- Thomas took off his pants in the meeting.
- Practice pronunciation – click once to start recording and again to stop
Next to the example, you have two buttons. The first one plays my audio version and the second one opens the recorder so you can practice.
To start recording, just click the big button once and read the example.
Click again to stop recording, and the AI will give you feedback how you did, both as a total and word for word.
You can instantly listen to your recording, as well.
You can try as many times as you want and compare your recordings.
To delete one, just press the red “X“.
To close the recorder, just press the small microphone button next to the example again. That’ll hide it.
Your takes will still be there, though, until you leave the page.
I think this way of doing it is much more intuitive and convenient to have it right there for each example, but you can just as well ignore it :).
So yeah… I still have to add a little “help” section for the whole thing, and currently it doesn’t handle examples that have “” in them properly. But I think it’s already much better than before.
And that’s just one step… because I actually want to give you all the option to hide the English or German version. So that means, eventually, you’ll be able to just see the English and then try to “speak” the correct German before you actually see it.
Not sure when I’ll do that, but yeah… it’s the plan.
Now, since there is no real help section yet, and we’re talking about the AI anyway, let’s do a quick FAQ for the feature.
Pronunciation Trainer – FAQ
The tool is not working doing anything.
The recorder needs permission to access your microphone. My site asks for that when you try to use it the first time, so maybe you denied then. To change that, just click on the little lock icon next to the website address. You can change permissions there.
You typically only need to do this once – not every time you visit.
Mic is enabled – but it’s still not working
Chances are you’re using a device from Apple. They’re not capable (or willing) of fixing well known issues with their Safari browser and some of their OSs. So if you’re using an Apple device please try with a different browser, like Chrome or Opera or Firefox. No guarantee, but for many people that fixed the issue.
What do the colors mean
I have set different thresholds, based on percentage.
This does NOT mean that you said it the way I said it. It just means that the AI grades this word as [xyz]% native speaker level.
I get very poor result
If you get poor results, it might be a problem with your recording quality. Listen back to one of your takes and check if the voice is loud and that there are now strong noises. Try getting closer to the mic if the voice is not very loud.
How good is the AI
The AI is doing pretty well at grading me (a native speaker), and it even understands all the every day stuff like omissions and shortenings of vowels and so on. I get very high grades even if I don’t make any effort to pronounce properly. So yeah… the AI does not need “super clean German”. It really grades native speaker style speech.
However, there are a few areas where it is struggling. Like… a few words where it’s a bit more tricky.
The “du” for example is one where I often get 60% and I need to take extra care and pronounce “more” than I usually would.
So yeah… it’s not a perfect system and by no means replaces feedback of a native speaker. But it’s pretty good at what it does and a nice way to get some speaking in :)
I’ll add more questions to this FAQ as (or if) they come up in the comments. But yeah… that’s my new feature.
Let me know in the comments what you think about it and if you’ve used the pronunciation trainer in the past. It’s no problem if you don’t use it. It’s really just an option, not an obligation.
But yeah, let me know all your feedback and also let me know if you have questions or if something is not working.
Next week, we’ll do a “normal” Word of the Day again. Till then, have a great time and learn some German :)
Brilliant idea, huge improvement.
In case it helps: on my system it fails. Using iphone 11 with blue tooth link to Phoned hearing aids (mic and speaker), accessing your site with the “Gmail -Email” app, and accepting “use microphone?”, I hear nothing, record silence, and get score of “NaN%”.
Then if I turn bluetooth off, I can hear your voice perfectly through the phone, but still record only silence. Opening your settings and turning mic to “denied” did not solve it for this time .
Unfortunately, at least from what I could gather so far, this is an iPhone issue that I can’t fix without significant hacking/coding.
It#s been an issue since 2017 and Apple doesn’t care to fix it.
It is such a great feature for self-learners. Also, I just got access to the Blog as sponsorship. I would like to express my deep gratitude to people who donate to help people in need like me. Without you, I will not be able to realize my goal.
Got it on my laptop, which is great, after being very disappointed in my Apple devices. Thanks – really impressed with this.
I had to switch to Chrome after opening the link in Gmail web view. And I had to enable for Chrome (ask every time option did not do it).
Samsung Android 12
And I got 89% score on the second attempt! Which is impressive given I only started in February
Ich habe 99% erreicht, deshalb bin ich sehr glücklich mit dein Upgrade! :-)
Das ist megagut!! Glückwunsch!!
With iPhone 16.3.1 it records nothing – under safari, Firefox or chrome. Each time I am asked to give permission to the microphone. Very sad.
I’ll do some searching again, but I have no device to test this on.
Does the button blink when you click it?
Endlich, funktionert es! Super!
Ich musste Google Chrome auf meinem Laptop statt auf meinem Handy (iPhone) verwenden. Danke für deine großartige Arbeit. Hoffentlich, werde ich endlich meine Aussprache verbessern!
Ja, viel Erfolg :)!
Danke! Übrigens, ist mein Laptop ein “Mac.” Ich bin so froh, dass es trotzdem funktionert
Für mich, funktioniert es nicht.
Any tips for a person who repeatedly grants microphone access but cannot record any voice input (the microphone doesn’t pick up on my voice). I’ve tried Safari and Google Chrome. Vielen Dank!
I’d love to help, but I don’t have any friends with an iPhone or Mac that I could ask.
Maybe Cari next time I see her.
But I’ll do some googling and I’ll ask ChatGPT about it. Maybe I can find something.
Sehr gut, es funktioniert für mich wohl.
Funktioniert GUT, sagt man. Nicht “wohl” :).
Mit wohl klingt es wie
Alles gut für mich. Für YDG benutze ich Chrome (auf einem neuen Mac) – und ich habe am drei anderen Mac Komputers überprüft. Am Samstag früh habe ich meine erste iTalki session. Wünsch mir Glück und Daumen drücken!
“Window 98” (Windows 98)
“all the option to hide” (all the options to hide)
“I get very poor result” (I get very poor results)
“replaces feedback of a native speaker” (replaces the feedback of a native speaker)
I haven’t tried the “speaking” stuff, mostly because the mic on my big laptop I have at home doesn’t work. I also have a Notebook, but that’s what I use in the coffee shop – it would be very weird if I started speaking German to a computer in front of all the other people…
But I’m going on a worcation this Saturday so I’ll try in in the hotel room :)
Works perfectly for me!
But I can’t get it to like my pronunciation of Meeting XD XD XD
I can’t get it to have an outline either – no matter whether I do German accent, or English accent or American accent.
What score do you get for it though? Should be at least green, right?
Oh yeah, consistently 98%.
That’s hardly “not liking” then :). Not sure what it needs to give you 100 but 98 is definitely good enough :D
I want that last 2%
I NEED it
Yes, the AI could give the the 100%. But maybe, it can give you something else that is much more valuable… Genügsamkeit.
You have so much… found God, build a great family with your partner, you travel, you help people. You probably put a tree in the ground somewhere (succulents do NOT count).
Maybe that last 2%… it’s okay not to have it.
Well, assuming that’s now meant to be live, I’d love to say that it looks amazing and works even more amaaaazingerlierly than before, but that would be a bit fat porky pie. Err, yeah, *lie*…
Well, the appearance side of the equation is 100% *swish/schick* (Wähle deine Sprache aus), but I never get a response from the AI. The big button glows red around the periphery, as per usual, stops doing so upon a second press, but no output shows.
This is using Google Chrome on my old iMac, which worked perfectly with the “U.B.E.” Windows 98 version. (And, continues to work: I just checked the “gehen” article, and the old interface still happily brings up the results.) Microphone permissions have been granted, I’ve quit the browser, force reloaded, “cleared browsing data” completely, but nothing has budged.
(1) Before clicking the big button
(2) After clicking it
And then we’re back to (1), with no feedback :-( I even cranked up “Microsoft Edge” and “Safari”, but they exactly mirrored the results obtained with Chrome. (But hey, I count it as progress that “Safari” at least asked for permission to use the microphone. Babyschritte, Babyschritte, Apple!)
For some reason, the container for the result got parsed out of the example.
It should work now, actually even without any proper script reloading.
Thanks for catching that so soon and for the screenshots.
Beautiful! Very happy to confirm that it’s a nice, compact bundle full of output-laden goodness now (at the end of the speech-input process, I mean). As per the screenshot below…
Great score!! I don’t do much better, at least not on phone without caring much about sound quality.