How Common Voice contributions have changed during the past month?

Hello Common Voice Community,

I hope you all are doing fine!

Due the current health crisis situation I know some of you have been asking about how Common Voice was doing and you have been also re-working your community approach to adapt to the current environment.

I want to share with you some numbers on the contributions during the last month (Mar 16 - April 15) compared with the previous period (Feb 14 - Mar 15, due Feb being a shorter month).

We see some languages with relevant increases in the number of voice recordings and validations, like Catalan, German, Basque, Frisian, French and Italian :muscle:

At the same time we also see some slight decrease on languages like English, Spanish or Chinese, and higher decreases on Indonesian, Japanese, Kabyle or Portuguese.

Please note these are percentages, so in languages with fewer contributions it’s easier to see higher variations than on the larger ones.

Thanks everyone for your contributions, keep the great work and stay safe! :slight_smile:

Language Records change Validations change
ab -100.00% -60.00%
ar 137.01% 60.86%
as 0.00% -28.13%
br -4.27% -10.86%
ca 212.11% 166.28%
cnh 2091.43% -44.83%
cs 291.00% -27.76%
cv -30.81% -4.72%
cy 655.82% 50.98%
de 147.60% 119.59%
dv N/A 1557.84%
el 72.75% 40.19%
en 13.25% -21.04%
eo 67.17% 21.15%
es -52.60% -35.46%
et -78.54% -98.50%
eu 169.35% 249.82%
fa -13.63% -7.45%
fr 17.01% 90.81%
fy-NL 434.14% 621.07%
ga-IE N/A 117.50%
hsb -4.80% -41.36%
ia -94.29% -54.74%
id -88.07% -86.94%
it 201.74% 87.71%
ja -91.94% -93.56%
ka N/A N/A
kab -51.02% -57.38%
ky -37.50% 383.33%
lv 2750.00% -54.79%
mn -88.33% -63.58%
mt 120020.00% N/A
nl 36.68% -3.24%
pl -43.21% -50.52%
pt -42.71% -48.26%
rm-sursilv 350.00% -14.68%
rm-vallader N/A 316.67%
ro 766.57% 623.39%
ru 9.64% -30.36%
rw 425.91% 7071.19%
sah -52.38% -94.29%
sl -44.44% -91.37%
sv-SE 65.98% 39.47%
ta 2017.31% 3931.05%
tr 1723.02% 1110.56%
tt 3331.25% 94.70%
uk 146.22% 129.88%
vot N/A N/A
zh-CN 30.88% -17.42%
zh-HK -1.29% N/A
zh-TW 3.92% 18.41%
7 Likes

Interesting. It makes me wonder what differences there are in the recording and validation contributors between communities. Is there any gathering of information about the who and why of contributors? Is that even useful information?

I don’t know if I quite understand the kind of information you are looking for.

You can get per language stats here: https://voice.mozilla.org/languages

We specifically don’t share individual contributors data to respect their privacy.

What are you looking for around the “why”? Is it possible to elaborate what are your trying to understand?

Thanks!

It does feel as if validators are a much smaller number than speakers. I remember before the clip priorities were changed, validation in English was 1 year behind recording and the gap has probably only grown since then.

I do feel as if validation is not as well-promoted on the site as it could be. I have suggested validation to some people who recorded and they have said things like “oh, that never even occurred to me.”

(I also think the line “some would say it’s the fun part” in the recording explanation suggests validation is not fun and may put people off validating.)

Recording is simple and does not require you to ask questions.

On the other hand, for validation, there doesn’t seem to be any FAQ that would answer a number of questions: should we keep a recording where a second person would speak in the background? Should we keep a recording of a person who is not a native speaker and who pronounces certain words strangely (and where certain pronunciations can make you think of other words)…

I think a lot of people don’t want to take the risk of making the wrong choices.

ps: sorry for my really poor english :frowning:

2 Likes

This is a great reflection on why we are getting a fewer validations vs recordings. Do you have the change to ask people who contribute to common voice to validate it? Maybe just asking:

  • Where do to spend more time on common voice?
  • Why do you spend more time here vs there?

Recording :
I repeatedly check a recorded phrase because,

  • I don’t want to record any lyrics other than mine (so I’ll start again if someone speaks in the background without knowing that there is a recording in progress, for privacy reasons),
  • I try to minimize the noise of the clicks to start and stop the recordings as much as possible,
  • I’m trying to keep up with the punctuation,
  • I start all over again every time I stumble over a word or stutter,
  • I try to keep the natural diction I would have during a discussion (no reading, sentences with missing numbers),
  • I’ll listen to it one last time before I send all five recordings.

Validation:
When I hesitate, I press the “pass” button (proper names, partial or total misunderstanding due to a handicap). I only validate the pronunciation if I consider it possible to understand the sentence in a maximum of 3 listenings without reading it.

I think this is accurate.

For me the recording learning curve has been about:

  • Paying attention on how I enunciate
  • Making sure I’m on Windows (Linux adds a weird noise to my microphone in Firefox)
  • Making sure no one around me is making any noise
  • Using keyboard shortcuts because my microphone captures the clicks of my mouse

And that is it. And even if you don’t care about it, odds are the recordings are good enough anyway.

But for listening, I agree with your comment. The website experience just throws you into validation without any help. And you hear audios with clips, with hums, with very low volume, different accents, etc. Personally sometimes I’m not sure if by saying “Yes” I may be making the system worse! And what if I say “No” and I throw away a valid volunteered effort! And the answer is to go to a Discourse forum and look for an appropriate thread that answers your questions. For a non-technical user, I think this is asking too much.

And even then sometimes the FAQs don’t help that much. In my case, I validate Spanish and the Spanish FAQ says you should say “No” to audios that skip the /s/ at the end of words. In my accent, it’s common to skip /s/ at the end of words, especially within some less economically privileged groups (Wikipedia puts it as “less educated” and I don’t really like that idea). So by rejecting audios without /s/ we’re making a system that leaves out the people that are already left out.

Anyway, it’s a complex issue and to be clear I don’t mean to belittle the efforts of a lot of people that created it as it is today. I think it’s a great project and that’s why I contribute :slight_smile:

Back on topic, living in quarantine with other people means I have fewer moments when I can be in silence and I won’t bother others. And validating-wise… well, not sure about other languages but in Spanish the top 3 people have so many more clips than the rest, that you hear them all the time. The first person in particular has 40k clips and the second has 24k clips, so the first person appears very often when validating. And either he has this really annoying and loud background “hummmmmm”, or his volume is really low. Sometimes I stop because I really don’t want to listen to him anymore lol.