🗣 Feedback: Common Voice Activity Bundles

Common Voice
#1

Update: We have published these activities with a better format over Mozilla Activate portal.

Hello everyone,

I’ve been working on series of activity bundles for Common Voice that I would like to share with you all for feedback.

This post is going to be long, so here you have my asks:

  • Test the activities by yourself first and then come back here to add your feedback.
  • What worked, what didn’t work?
  • Are we missing anything?
  • Is there another scenario you think we can design an activity? Please describe it.

My goal is that we can iterate and improve these activities during March by testing them on the ground so by the beginning of April we can post them as official activity recommendations and actively promote and encourage people to do them.

Purpose

In order for a dataset to be useful, it needs at least 2000 hours of validated voice recorded by at least 1000 different voices. Voice diversity is key: gender, age, background noise, accents


Guiding principles

We believe we can get this voice diversity from communities participation, and we would like to see around 10% of the total goal (around 200 hours) in a few locales by the end of June.

Our calculations tell us that we can get there by engaging 800 people in each language, each of them donating 15 minutes of voice (225 clips) and helping review 450 clips. In order for this to work we need at least 180 000 sentences available on the site. These numbers are optimized to avoid contributor fatigue, we assume some people will donate more and some less.

We also understand that for these donations to be engaging we should design for it, having in mind different scenarios where these donations can take place with different audiences and time availability.

Also, we want to make sure people have a basic understanding on the importance of this project to the world, user’s privacy, open innovation and the value it can unlock to everyone. This will increase engagement.

Participants can also unleash a viral effect to get more people involved and promote this project, word-of-mouth and social media.

Getting people to create an account on the site and subscribe to the newsletter is something we really want, as a way to be able to follow-up with them later.

Audiences

  • Individual contributors: People interested in contributing from home on their own, at their own pace.
  • Small groups: Close friends or family groups that want to see in action what this project is about.
  • Medium-large groups: Students, event attendees.

Proposed bundles

:cowboy_hat_face: Lone cowboy/cowgirl

  • Audience: Individual contribution
  • Time needed: From few minutes to 1 hour (total)
  • Outcomes: 225 clips recorded, 450 clips reviewed, optional social sharing, follow-up to next bundle, subscribe to the site.
  • Material: Computer or smartphone, microphone

Visit https://voice.mozilla.org/ and create an account, make sure you subscribe to newsletter updates.

  1. Click on Speak
  2. Check the sentence on the screen
  3. Click on the red microphone icon
  4. Read the sentence out loud
  5. Click the red stop icon
  6. Repeat until you have recorded at least 5 sentences
  7. Click send and repeat

Now it’s a good moment to share this if you like on your social media:

Help teach machines how real people speak, I’ve just donated my voice at https://voice.mozilla.org #commonvoice

Donating 15 minutes of your voice is enough, this is more less 225 clips. Once you have reached this number it’s more useful if you to to the Listen section and help review other people’s voices.

  1. Click on Listen
  2. Click Play
  3. Listen to the voice (try not to read the screen)
  4. Now read the screen and compare with what you listened
  5. If it’s the exact same click yes, if not click no (note different accents are fine)
  6. Repeat

Ideally each person should reach at least 450 clip reviews, feel free to devote a few minutes each day.

Once (or as you are getting there) you have recorded 225 clips and reviewed at least 450, it’s probably time for you to go to the next challenge, check the “Fun with friends” activity.

:family_man_woman_girl_boy: Fun with friends

  • Audience: Small groups (3 to 10)
  • Time needed: 30 minutes
  • Outcomes: 15 min recording, 15 minutes reviewing, optional social sharing, follow-up with 3-5 people each, subscribe to the site.
  • Material: One smartphone per person, ideally headphones with microphone

Gather together 3 to 10 friends or family. Briefly talk with them the concerns of the current voice-recognition systems and the privacy implications. Getting a few minutes so everyone can talk about their reactions and concerns is a good idea.

Intro them to the Common Voice project, why it’s important, show them how easy is with just a smartphone to donate your voice and to review other people’s voices.

Ask them to use their smartphones and drive them through how to create an account. Spend 15 minutes and have fun recording voices.

After that, spend 15 minutes reviewing other people’s voices (enjoy other people’s accents and voices!)

Invite them to talk about this in their social media accounts:

Help teach machines how real people speak, I’ve just donated my voice at https://voice.mozilla.org #commonvoice

Invite them that in the following days they run this same activity with 2-3 people they know and share the experience back with you.

:woman_juggling: Show time :man_juggling:

  • Audience: Medium-large (10-100)
  • Time needed: 1 hour
  • Outcomes: Intro to the project (10m), 15 min voice recordings, 30 minutes clip reviews, optional social sharing, follow-up each each participant to run “Fun with friends”, subscribe to the site.
  • Material: Projector, poster, slides, computer, one smartphone per person, headphones with microphone highly encouraged

This activity is divided into four activities.

1. Intro to the project (10 minutes)

The activity owner will run a max. 10 minutes presentation (slides you can use) to introduce the project to the audience, the key elements that this should answer are:

  • What’s this project about?
  • Why it’s important?
  • What are our goals?
  • How can I be involved?

2. Voice donations (15 minutes)

Individuals will sign-in the site and with their smartphones and microphones they will start donating their voices for 15 minutes. Activity owner will control the time and ask people to move to the next activity.

3. Review time! (30 minutes)

People will make groups of two people and will go to the listen section of the site. One person of the groups will play the recording on his/her device and the other person will just listen (without reading the screen). The person listening will repeat what he/she has heard, if it’s the same as the screen is showing the first person will click “Yes” and move to the next one.

Activity owner will control time and ask for couples to change roles after the first 15 minutes.

4. Wrap-up and sharing (5 minutes)

The activity owner will ask participants to optionally share on their social media about this activity

Help teach machines how real people speak, I’ve just donated my voice at https://voice.mozilla.org #commonvoice

People will be also asked to make sure they subscribed to the Common Voice newsletter and check the “Fun with friends activity” as something easy to do in the next few days.

:speaking_head: Reach the crowd :star_struck:

  • Audience: Medium-large (10-*000)
  • Time needed: From 1h
  • Outcomes: Get random people to donate 2-3 minutes of their time at one fixed place where crowds pass-by (like a booth), optional social sharing, follow-up with fun with friends, subscribe to the site.
  • Material: A few laptops or tablets with microphones, ideally with headphones.

A set of at least 2 laptop/tablets will be set up in the fixed place with only Firefox in full-screen mode with https://voice.mozilla.org loaded.

Ideally a paper next to the device will read:

Machines should understand your voice without compromising your privacy.

Donate 3 minutes of your voice now to help [LANGUAGE]!

  1. Click on Speak
  2. Check the sentence on the screen
  3. Click on the red microphone icon
  4. Read the sentence out loud
  5. Click the red stop icon
  6. Repeat until you have recorded at least 5 sentences
  7. Click send

Ask anyone in this place if you need assistance or more information.

People in the fixed place should encourage people passing by to know about how they can help protect their privacy if they donate 2-3 minutes of their time.

Examples:

  • Hello! How are you doing? Do you know you can help machines to understand you and protect your privacy?
  • Hi! Did you know Common Voice wants to teach machines to understand us and protect our privacy?
  • Hello there! Would you like machines to understand your voice without sending it to Google or Amazon?

After this initial approach we can quickly introduce people to the Common Voice project and invite them to donate 3 minutes of their time now to help their language be understood by machines in a privacy-aware way.

4 Likes
Gender bias in voice tech & personal intelligent assistants
Share your event printing materials
#2

I guess the question is how do you have 20 people in a room recording at the same time without their voices interrupting others’ recordings?

Also, with regards to validation, you bring people together in a room and then they all put on headphones and ignore each other. Perhaps this could be made more engaging with a social or competitive aspect, like who can validate the most in x mins?

Perhaps the site can encourage this with group leaderboards. So you add your friends to a group and it shows your rank (number of accurate clips) relative to your friends. The site could make sure that there is some level of clip overlap between group participants so you get accuracy information immediately.

Having some kind of group functionality on the site makes it so that you can still do these kinds of activities with your friends even if it’s not practical to get everyone in the same room at the same time.

#3

This is something we’ll have to test on the ground to form an opinion. I suspect phone microphones are good enough to capture the main voice, but we’ll have to test it.

Definitely a good idea, let’s try it and report back the experience.

This would be great, and we should keep it in the idea box, but since it involves app changes we can’t expect this to happen before the end of June. That’s why I’ve focused on things we can do with the tools we have today.

Thanks for your feedback!

#4
#5

Hello again,

Did anyone have the chance to test any of the activities on the ground? We would love to know how did it go :smiley:

Note that we also plan to test these activities with local Mozilla communities in April, so all feedback will be helpful to improve them.

Cheers.

#6

‘Individual Contribution’ and ‘Fun with friends’ category is perfectly defined. We can, in a room, with 7-10 people, record and donate voice without any noise in the background. But if we do it with 10+ people, we need to have the audience divided into groups with a separate room/section for each group.
Besides, it would be great if we can design a generic poster or something for the activity so that it becomes easy for the organiser to spread the word. Moreover, we can have a presentation prepared which the organisers can use to make attendees understand what the project is all about and why they should contribute.

#7

What has been your experience here? When did you have to split people? Groups of 10?

#8

Sorry, I forgot to update, the first post now has the slides and poster.

#9

Hi again,

We are still looking for your experiences with these activities!

Did you have the chance to try some of them? What do you think?

Thanks for your time! :slight_smile:

#10
#11

Hi @nukeador, can we have the original pptx file of the slide? (perhaps just zip and upload here). When I try to download it from Google Spreadsheet it’s been convert “into” pptx again, and it seems cause some bad format.

#12

Hey Irvin,

Here’s the deck linked from the Activate activity.

#13

Hi @lucyeoh, this is the one which had already been converted into “Google Spreadsheet” format, and while we downloading, it will convert into pptx again, resulting some weird glitch.

It would be handy if we can have the pptx or zip file direct from Google Drive, but not from Spreadsheet.