Show HN: Leova – An API to Add AI-Powered Voice to Mobile Apps in Mere Minutes | Hacker News Reader

Show HN: Leova – An API to Add AI-Powered Voice to Mobile Apps in Mere Minutes

leova.io

19 pointsdigital_ins13 comments

Hi Everybody! I'm part of the team and we built Leova to make adding fully-functional natural language interfaces to your home-brewn or production-scale apps would be super easy. Integration is (supposed to be) super simple and deployment, even easier. We've got a couple of short videos on the website to show off how effortless it's supposed to be.

The NLP AI is conversational and as people use it, a small machine-learner learns to 'fix' errors in the recognition of whichever speech recognizer it has been paired with it. The entire system has been modeled around human conversation and works without keywords and without forcing users to speak to it in a rigid way, like an IVR system.

This first release is for LeovaTravel, which is this one :-)

The next 3 months will see us deploy the next two variants - a natural language AI for ordering food; and one for shopping / ecommerce, both wrapped in APIs.

This would actually be a lot better with speech recognition built in. Is that something in the works?

Actually, today's speech recognizers are really good and ones like Google's have accuracy >90%. So writing a speech recognizer ourselves with 50% accuracy and training it endlessly without ever reaching Google SR's accuracy sounded like a terrible 'reinvent-the-wheel' idea.

Instead, we wrote algorithms that further boosted the accuracy of the existing speech recognizer with machine-generated phonetic dictionaries. (You can think of it as standing on the shoulders of giants). The plan is to work with a couple of partners to train our machine learning system for their products to really bump up SR accuracy.

If you're building an app that's going to go live, send us a message, we'd be happy to work with you (for free!)

Maybe it's just me ... or does the demo seem to be sped up? Like soundhound's demo...

I'm a part of the team and lol, no - that's just Chadd's "demo voice". I think the feeling that the demo was sped up comes because we made those card transitions so fast. If you sign up, you can actually try the demo yourself to see that it really is that quick :-)

orrrrrr, if you watch the demo video really closely, you'll see how you can try the system out without actually having to sign up

I see you are using google speech recognizer for chrome and I guess it works the same with android, how do I use it for developing apps for iOS. Does it work with Siri ?

For iOS development you can use iSpeech Recognition, OpenEars, Nuance or CMU Sphinx for speech to text. We use iSpeech and it was really simple to integrate.

Ya but ispeech accuracy is not that great

So that's where Leova really shines. We saw this shortfall in accuracy when we were using iSpeech for testing on iOS, so we built a machine learning system that generated structured phonetic dictionaries.

This means that Leova provides a layer of error correction over and above specifically iSpeech's transcriptions. And with frequent usage (days, not weeks) you'll see Leova's interpretations of buggy iSpeech transcriptions really blow competing NLP parsers' resilience out of the water

if you're curious to try this out - I'd recommend you sign up (free), download the IPA file (also for free) and try it out (also free!).

To help speed up your ios dev, we've also posted all the libraries you'd need on GitHub and on the first page of the Leova admin console, once you've signed in

Seems to be pretty accurate - Is nltk being used on the backend?

Thanks! We've actually built our own probabilistic information extractor, so we don't use the POS tagging, so NLTK is sorta redundant for the approach we took (and also less accurate)

Reply on news.ycombinator.com