Common Voice - Donate your voice to teach machines how people speak | Mozilla

It’s frustrating when you’re not understood — especially when you’re trying to speak to Siri, Alexa, or another internet-connected device.

Voice datasets that power voice recognition services are owned by a handful of major companies, and they can wildly underrepresent the voices of non-dominant accents, Black, Indigenous, and other people of color, disabled people and gender marginalised people. In fact, for people speaking other global languages - there may be no datasets at all.

That’s why Mozilla launched Common Voice — the world’s largest public voice database, powered by the voices of volunteer contributors. Our goal is to teach machines how real people speak.

Today, we’re asking you to contribute to Common Voice, but we want you to choose how you’ll do it. Will you donate your voice to one of our Common Voice language datasets? Or will you make a $34 donation to Mozilla to support projects like this to reclaim the internet? (Or both!)

I’d be curious about the privacy concerns, but this might help a lot with underrepresented voice data. It might come down to if someone wants more datasets for their particular voice/language more than the other concerns.

If your language/accent is already well documented, it might not help as much?


Nice try, Ursula


Why would black or gender marginalised people have a different voice?


Because the genetics that build the vocal tract could be different, which in simple terms could mean a change in pitch. There are also more cultural differences such as speech cadence, accent, and inflection.

Appoxo, avatar

Have you seen how (even if stereotypical) black folks talk vs white folks (in media)?
Just turn on GTA 5 and listen a bit to Lamar and Franklin talking in missions.
So much slang is in there.


Right? Like we all are made the same on the inside


Dude, I clicked on the link pretty excited to volunteer. I have a professional mic, a little time, and a decent voice. The first thing that greets me is “Voice datasets also underrepresent: non-English speakers, people of colour, disabled people, women and LGBTQIA+ people.”

Well, I’m none of those. So maybe they don’t want my donation, or I’d spend time and they wouldn’t use my recordings… Sort of a letdown.


Technically there are different dialects and a lot of unique slang, idioms and specific descriptive words.

In the trans and non-binary community for instance there’s a lot of terms regarding how people identify and express themselves that unless you know the actual function of how they work aren’t easily indistinguishable from slurs to outsiders. Take “Femboy” and (please forgive me mods) “Shemale”. The former is a perfectly socially acceptable description of a guy (cis or otherwise) whose gender expression is very feminine…the latter is a slur that places emphasis on the birth sex characteristics of a trans woman and implies heavily they are guys just pretending to be women and the term originates from the porn industry that fetishizes trans women.

You also have the usage of neo-pronouns. In languages with more gendered components than English sometimes what words are chosen either reflects the gender of the speaker or the person being addressed or objects can be given a gendered connotation. Some languages are actually very gendered and the usage non-binary folk using those languages make whole new conventions. English speakers whine a remarkable amount over they/them singular pronouns are confusing but ain’t seen nothing. A lot of places your job title and status has no neutral gendered term or culturally there are sentence structures that differ down entirely binary gender lines. Are you latino or latina? Guess we need a new word… Latinx!

FaceDeer avatar

Mozilla: "We'd like to build a dataset of underrepresented languages and accents so that voice recognition works for everyone. It'll be under an open license."

Most of this thread: "GIVE ME MONEY."

Sigh. As soon as it turned out that AI training data was "worth something" everyone turned into a money-grubbing mercenary.

gedaliyah, avatar

I’m not sure why there is so much anti-Mozilla hate. I know they’re far from perfect but they do an awful lot for the open source world. Having an open database for voice training seems like something that the world can use to do some good.


Because everyone knows better how to do open source, and these ppl are usually right ideally, but when you apply some concepts you can starve to death. Much of what mozilla does are not ideal, but are very good, and the only option of things we need today, not in 20-30 years.


The problem with mozilla is they forget about their browser and completely ignore it. It’s almost like even they have given up on firefox…


Because they are shady, extremely profitable and yet ask for donations like they are some greater good opensource nonprofit startup.


gedaliyah, avatar

But when I donate my voice, it’s not going to some vault at Mozilla. It becomes part of an open resource that anyone can use to build models, libraries, etc.

Just because it is organized by a company that may or may not have nefarious goals, isn’t that still a good thing to exist?


Let me completely exaggerate to illustrate the concept:

If osama bin laden or hitler, mao, a terrorist org etc. start a charity to plant more trees you would feel uncomfortable planting trees for their charity.

If I don’t fully trust a company, it discourages me from participating in anything they do, no matter the intention.

RobotToaster, avatar

Mozilla pay their CEO $5m a year, yet they want donations.


Not today skynet

auf, avatar

As long as the actual software would be free and open-sourced, I’m willing to help

yo_scottie_oh, (edited )

The data set is available under the Mozilla Public License v2 through the Common Voice GitHub page. I’m not sure if I’m reading the terms of the license correctly, but I believe it allows commercial use.

otter, avatar

I think that might be a part of the focus, to push companies into including these underrepresented languages/accents so that the products work for everyone instead of a smaller subset

Worth considering before contributing

gedaliyah, avatar

I’ve been using this CV Project app for months whenever I have a few minutes.


Every free and open-source license allows commercial use.


I used the data set for my bachelor thesis. Thank you, Mozilla!

auf, avatar

bruh the license sucks. Not gonna do that until they change it

sir_reginald, avatar

what’s wrong with the MPL? it’s a pretty standard free software license. It’s more permissive than the GPL, but less than the MIT license. The most relevant project using it is Firefox, but you can go to github and find a ton of projects using the MPL as their license.


Google pays them $450M/year and they have the audacity to ask for a donation?

No. I don’t think I will.


Why Google pays these money to Mozilla?


Well it’s primarily to have Google as the default search.

But I assume it’s also because it’s their nearest competition and the situation is so bad that they’re trying to avoid anti-trust concerns.



  • Loading...
  • FaceDeer,
    FaceDeer avatar

    So, wait, you're fine with Mozilla impersonating you as long as you get a little money in the process?

    Not that this is what Mozilla wants this data for, mind you, I'm just puzzled by this place you've jumped to.



  • Loading...
  • FaceDeer,
    FaceDeer avatar
  • All
  • Subscribed
  • Moderated
  • Favorites
  • DreamBathrooms
  • InstantRegret
  • osvaldo12
  • magazineikmin
  • rosin
  • Youngstown
  • mdbf
  • everett
  • slotface
  • khanakhh
  • thenastyranch
  • kavyap
  • Durango
  • rhentai
  • bokunoheroacademia
  • tester
  • Leos
  • cubers
  • ethstaker
  • GTA5RPClips
  • modclub
  • tacticalgear
  • relationshipadvice
  • lostlight
  • cisconetworking
  • normalnudes
  • HellsKitchen
  • sketchdaily
  • All magazines