Then you need to do what’s called alignment, which is chunking it up into sentences and have those sentences transcribed, like breaking this down into small pieces. We can help with the verification of that if that was something you would like. We can get this in front of people, ...

In fact, over the last few weeks, because there was news here about Common Voice, we’ve had a lot of Taiwanese users. If we could take some of the text from this and have people read it, if you want, we can also validate some of the transcription. As part ...

This is something we would love to work with you on. This data set, that four to five hours of traditional Chinese with Mandarin, is a very valuable data set. We can do a number of things to help with this. One thing that Common Voice does is, as Des ...

We would love to talk to you about these databases and how we can work together to create the databases. We already have quite significant infrastructure for doing such things. We could talk about mirroring or how we could work together from a technical perspective. That would be easy for ...

CC0 or just public domain is what we publish. We would love to help with the recording and preservation of languages. Of course, this is a big goal of ours. It’s not just cultural preservation through technology, but also speech technology. We think people should be able to speak to ...

First of all, the way we see Common Voice and Deep Speech, we don’t see this as a tool for Mozilla to go out and collect language data. We see this as a tool to allow communities to build up this data set for themselves, and then use this data ...

I have some ideas on stuff we’d love to collaborate. Before I go into that, are there any other questions off the top of your head? If it comes up, we can talk more. There are a couple things that we would love to collaborate with you, something we can ...

Excellent.

Centralized database.

One thing I would love to hear a little bit more about are these databases that you will set up through the Indigenous Languages Act. Can you give us some information about what those databases look like? It seems like we could collaborate on that.

It sounds like a hugely valuable data set to at least bootstrap research projects or start up projects. From that, you can continue to build upon that data set. We’d love to use something like that as a base, but it’s not necessary.

We’d love to talk about that.

I would imagine this data set of four to five thousand hours of traditional Chinese Mandarin is not a public domain, licensable data set.

...conversational, I guess.

Are they...?

That’s awesome. Can you tell us more about this data set, like how it was collected and curated? What’s the licensing? Do you know much about the 4,000 to 5,000?

...speech.

So NLP stuff...

Like that or not like that?

What kind of goals are they setting?

It’s nice to hear.

Wow.

Oh, they own the IP.

Or is it all social good?

As for output from some of these challenges, is commercial output viable? Can you build something that could make you money, to be a commercial entity?

When is the challenge?

Which one is it?

OK.

We have some things that we’d also like to speak about, perhaps, but we’d love to hear from you.

That’s why we’re doing, make it easy and free for anyone to create this speech technology. That’s our goal. That’s us. That’s what we’re working on. Common Voice is data. Deep Speech is the technology, software, platform. We’d love to hear more about what you’re working on and maybe what ...

You need, first of all, this data piece that we’re talking about. You also need the algorithmic machine learning scientists, people like Kelly. You also need very expensive compute resources. This prevents a lot of people from taking part in this technology. We hope to provide the software piece, so ...

Right now, if you look at speech technology, and if you think about the forefront of that, which is personal assistants like Siri, Alexa, Google Home, it’s only a handful of big corporations. It’s hard, because it’s only these corporations because it’s very hard to create speech technology.

One last kind of visionary piece, and then we’ll open it. I love to hear about what you’re working on, what some of your goals are. Maybe there’s some collaboration points. As Kelly and Dees mentioned, our goal is to open the speech and language type-wide for creators, hobbyists, universities, ...

Once we have the data, we also hope to provide software and tools for people to build speech technology with. Maybe that’s a good segue for Kelly to talk about Deep Speech.

We will add profiles to the this, so people can have identity, optionally. Of course, we will always accept anonymous donations. We want to expose a lot of metrics onto our site, so people can see how the project is progressing, so people can see how fast different languages are ...

We’re adding new ones all the time. The website, we currently have 50 more languages in progress. We’re continually trying to add to our collection of languages. That part of our effort is multiple languages, expanding the data set. Also, part of our effort for the next six months is ...

We published our first version of the data set in November last year, which was about 500 hours of English. Since then, we’ve about doubled that. We also started to collect in multiple languages. We’re currently collecting in 15 languages, big ones and small ones. The big ones are English, ...

People can also listen to those recordings and verify if what they said is accurate. This is also an important part of the data set collection, because we needed to be very accurate if the speech technology part is going to work. That’s the whole project. Currently, we’ve collected about ...

Also for things like language and cultural preservation for under-resourced languages, which is a use case that has recently emerged with languages like Welsh, and other smaller languages. It’s very simple how it works. If you haven’t seen it, it’s just the website where people can go and read sentences ...

Common Voice is Mozilla’s initiative to crowdsource a large data set of human voices in every language, every accent, every dialect, every gender, to be used in speech technology, specifically speech-to-text technology, but other forms of speech technology, as well.

I can give a little rundown of what Common Voice is and what we’re trying to do over the next six months. Maybe that would be a good place to start.

Common Voice was his idea. He’s been...

Common Voice is a part of Open Innovation.

Nice to meet you. My name’s Michael Henretty. I’m currently leading a Project Common Voice for Mozilla.

A process of checkbooks, got it.

So you don’t have a checkbook or you do have a checkbook?