Possible software for children to study Chinese.

[roddy]

Hey, less than a decade ago Skritter were apply for $25k in community college funding. Those guys went on to Codecombat and way more zeros. Good luck!

 

Also, though I hate to say it, maybe you shouldn't get fixated on Chinese. Might be easier to just start with English or something. 

[thechamp]

The google speech recognition service doesn't even offer Chinese....I think it must be really quite hard if that's the case. I can't find it on AWS either. It maybe even these guys are struggling to get effective speech to text software working with Chinese.

 

Also the processing for AWS Speech to Text service is NUTS! $4:00 for every 1,000 requests. At those rates you basically have to build your own speech recognition software.

[imron]

54 minutes ago, thechamp said:

It's quite a big ask.

Not that big, especially considering the problem space for this is significantly smaller in terms of what it needs to do and understand.

[thechamp]

Baidu have cracked it and they've released the model via github but there's no public API for the service....and without access to masses of data on Chinese speech-to-text stuff, it's a bit meaningless. There are existing datasets of Chinese speech to text.....but you also need examples of learners approximating the language to varying degrees of accuracy.

[imron]

1 hour ago, thechamp said:

Also the processing for AWS Speech to Text service is NUTS! $4:00 for every 1,000 requests.

These guys do speech to text, including for Chinese and at the most expensive pricing option it's $1.22 for 1 hour of audio.  Wit.ai also does speech to text and is free.

27 minutes ago, thechamp said:

but you also need examples of learners approximating the language to varying degrees of accuracy.

If it was easy everyone would do it, and that's why there are so many Chinese learning games for beginners.

[roddy]

If you've got four people and a year, spend a week looking at some big ideas. Even if they don't come off you'll get plenty of smaller ideas out of it. And as you say, transferable skills...

[thechamp]

Interestingly this has just been on hacker news: https://segment.com/blog/spotting-a-million-dollars-in-your-aws-account/

 

I just don't know how possible collecting this kind of data is. Perhaps it's the sort of thing you'd argue you'd do in later developments. It's not really a technology problem it's just a case of not having a dataset on beginner/intermediate/advanced learners. (Native datasets exist though).

Yes there is a lot of scope for thinking big, but ultimately we all want to get something working after one year. You could build the software on native speakers, to demonstrate it I guess. 

We'll see what my eventual team thinks. Also the incubator themselves have identified business areas that these cloud services are opening up possibilities in, so there are ideas on the table from the get-go. They're quite anxious to make sure the tension between their ideas, and each individual team member's ideas are well-balanced though. Everyone is supposed to get a fair say in what they want to do then we all have to agree. Obviously I quite like the look of the language learning market, but as you say - it's saturated.

[mikelove]

Pretty sure WaiChinese already tried the human tone checking idea. (not sure if they're still around, haven't updated their FB / Twitter pages in a year and a half - might be a cautionary example for an AI startup on this, actually, since while your costs at scale would of course be lower, your initial development costs would be considerably higher)

 

Voice recognition services for Chinese actually don't seem to work well for pronunciation checking in our experience because they rely heavily on context - i.e. they'll accept really bad tones if there's an obvious word with that combination of syllables, they're not trying to recognize syllables in isolation. (we see a similar problem with OCR - it'll pick say the 4th best match for a particular character because it's confident that the character next to it is such-and-such and so it makes sense for that 4th best match to be there)

 

If you're not hell-bent on Chinese, my go-to startup suggestion would be "Pleco for something other than Chinese" - the dictionary app situation for most other languages is quite poor and there's a very good business to be had doing basically just what we've done but for Spanish or French or German or Korean (or Japanese, but that's the only language that has something like a comparable app/website constellation to Chinese, plus like Chinese it has a large assortment of open-source dictionaries that make the bar to entry for new apps low). Or, if you can manage it, Arabic - come up with a better reference for Arabic and the world's defense ministries will show up at your office with dump trucks full of money. Reference publishers seem to finally be getting past their brief phase of insisting that everything be done within apps that they control, so you might find that the royalty advance for a good dictionary of Spanish or whatever is well within an incubator startup's budget.

 

Frankly, and I don't just say this for anti-competitive reasons, if you're going to do a language learning startup Chinese is pretty much the worst possible language to choose, because while the market is big it's also very very crowded and it's hard to rise above the fray - for any given Chinese learning product idea, there are probably a dozen other companies doing the same thing, and most of them aren't making any money. Even if you come up with something genuinely innovative, you still have to find a way to rise above the noise of the many other companies doing something innovative with Chinese.

 

On the other hand, with a more neglected language, the market may be smaller but it's also much much easier to get people interested in what you're doing; you can always expand that to Chinese later once you've got a larger business and a set of proven designs / concepts, but chasing after Chinese from the get-go makes it much harder for you to even reach that point.

[thechamp]

Thanks. Don't worry it was just an idea, and I still have 3 other people to convince of any idea, and also the people representing the incubator's stake (i.e. it's unlikely I'll be able to make the claim we can get AI for pronunciation grading working in one year). Although, like I said, amazingly transferable skills from something like that. One major reason for me doing this in fact is building an interesting product from end-to-end. If it fails I'm really not too bothered but I'd just like to have worked on something interesting for that long.

 

It's just language learning is an interest of mine so thought quite quickly of that. Thanks for the advice though!

[Shelley]

I use HelloChinese because I think it has very good speech pronunciation detection. I also agree the market is over run with chinese learning apps for children.

 

I find it hard to think of something different from all the things already out there. I suppose you just have look at what's there and pick the things that no one does well and improve it enough to actually be noticeably better.

 

What ever you choose to do, I hope it goes well for you.

[Lumbering Ox]

3 hours ago, mikelove said:

If you're not hell-bent on Chinese, my go-to startup suggestion would be "Pleco for something other than Chinese"

 

Maybe I missed something but I've looked at Japanese a bit and if you could do a Japanese Pleco I think the weebs would go sex nuts and retard strong to quote Clerks 2.

 

I am not so sure about other languages though. Pretty sure something like French and German would have a choice of Plecoish options and unlike Japanese or Chinese it is pretty trivial to type in the word you are interested in.

 

Smaller market but I swear by the many arms of Vishnu that maybe Hindi would be a good idea.

 

I seem to recall that Japanese has a program that allows you to point at a word you are reading and it gives you the pop up definition and IIRC you can add it to an Anki deck simply. I don't think they have one for Chinese, that would be total steak sauce.

[mikelove]

1 hour ago, Lumbering Ox said:

I seem to recall that Japanese has a program that allows you to point at a word you are reading and it gives you the pop up definition and IIRC you can add it to an Anki deck simply. I don't think they have one for Chinese, that would be total steak sauce.

 

We have that in Pleco, actually, assuming you're on Android. Is this pointing at a word on your phone's screen, or pointing at a word in a printed book with a camera? With the former you'd use our free Screen Reader function, with the latter you'd need our paid Optical Character Recognizer add-on; in either case the Anki integration is free (via the AnkiDroid API).

[Lumbering Ox]

1 hour ago, mikelove said:

We have that in Pleco, actually, assuming you're on Android. Is this pointing at a word on your phone's screen, or pointing at a word in a printed book with a camera? With the former you'd use our free Screen Reader function, with the latter you'd need our paid Optical Character Recognizer add-on; in either case the Anki integration is free (via the AnkiDroid API).

 

The Japanese version IIRC works on the computer screen. I think only on the computer screen.

[Guest realmayo]

Interesting that it's hard for software to judge you if your tones pass or fail when saying different words.

But why not do the judging ourselves, and instead use a piece of software to simply tell us where we sounded different from a given piece of audio.

 

So: you give the programme an mp3 of a sentence spoken by a native speaker, which you want to reproduce. Then you record yourself saying the sentence. Then you get visual feedback on where the differences were.

 

So if you ask it to give you feedback on the tones: you get a waveform of the original, and of your own input, one above the other, with the pitch changes marked.

Do that for all the variables: pitch, pitch changes, speed, consonants, vowels.

 

Free software -- like Praat -- can do the analysis already. But I'd pay for something which provided lots of extremely user-friendly tools for manipulating that analysis, e.g. zeroing in on a short amount of audio, letting me re-record a second attempt, letting me see how that second attempt compares with my first attempt as well as with the original mp3 recording, that kind of thing. Archive results so I can see changes over several weeks. And so on.

 

Okay, this wouldn't be for children but for adults. I'd buy it. If you gave me a discount.

[Shelley]

I think realmayo's idea has merit, and I do think you could adapt it for kids.

 

If there is a visual of the spoken word/sentence and they have to speak it and make it match the one they can see it might well help with pronunciation. You put the original in one colour and their voice in another colour and overlay them to see how close it is, I think it would be relatively easy to write some code to check if these two things match or not.

Bright simple graphics could be used to show the sound to follow, rewards at an appropriate level for how close they got. You could follow this through with the other aspects of learning chinese, follow with your finger characters, simple flashcards and sentence building (move blocks with characters/pinyin to the correct order) incorporating grammar.

 

The more advanced things mentioned can be in the "advance" tab for the adults to change/use with their kids, I don't see why parents shouldn't be involved and when the kids go to bed the adults can check progress and see what needs improving and make the required changes for the next days lessons. Or they could turn it in to adult mode and use it themselves??!!

 

I would be inclined to have more comprehensive grammar etc notes so adults could help explain if the child was finding something difficult.

 

This is one thing I haven't seen in kids apps for learning chinese, but of course I haven' seen them all. It might be worth a look at the best sellers see what they have.

 

Oh and yes I am available for consulting work......just kidding  Busy as .... these days.

 

 

[Guest realmayo]

8 minutes ago, Shelley said:

I think realmayo's idea has merit

 

That's nothing. Wait til you hear the name of the programme:

 

Shadow Shifu.

 

[Shelley]

Was thinking.. this could be your USP - family software, an app for the whole family, sign up each member of the family with their name and age then everyone could have their turn at a level appropriate for them. You could make a feature out of the sharing, family aspect.

 

From a money point of view a family subscription/single purchase how ever you do it, a bit more than the price for one, so more money in.

 

 

 

[艾墨本]

I want Realmayo's idea. 

 

I want to be able to take a recording that a Chinese friend does for me of a sentence or two, plug it into the software, and repeatedly practice imitating what the native speaker said. I want to be able to do this easily and hear back what I just said. Visual feedback of the difference between what I said and the native speaker would be great. Tools like what Shelley listed would be great, too. Being able to do this with a wide variety of audio formats would be really great. I also want to be able to change the speed of the audio playback without it distorting too much. And for the paid subscriber, include audio and dialogue samples for the user to use so they don't have to pester their Chinese friends to create recordings. A bonus would be automated feedback on where what I said and the native speaker said are different.

 

Please.

[zander1]

2 hours ago, 艾墨本 said:

Please.

 

Seconded.

[lips]

2 hours ago, 艾墨本 said:

I want to be able to take a recording that a Chinese friend does for me of a sentence or two, plug it into the software …… automated feedback on where what I said and the native speaker said are different.

Quote

software for children to study Chinese

?