Open Speech Recording | Hacker News Reader

Open Speech Recording

aiyprojects.withgoogle.com

45 pointsjonbaer18 comments

Folks interested in this should contribute to Mozilla instead: https://voice.mozilla.org/

Why help an ad company when you can help a constant warrior for privacy, freedom, and open source technology? Why is Google trying to create their own instead of joining Mozilla's work?

Not defending, but maybe explaining...

A HN headline yesterday explained which licenses were prohibited by Google:

https://opensource.google.com/docs/thirdparty/licenses/#bann...

Most of the licenses on the Mozilla page are either restricted or banned at Google, based on a cursory crosscheck.

https://voice.mozilla.org/en/data

Note that the other licenses listed lower on that page are for other data sets. Mozilla's data is CC0, which shouldn't be a problem for Google, because it doesn't include the "noncommercial" part.

Text of CC0 for discussion purposes, is here: https://creativecommons.org/publicdomain/zero/1.0/legalcode

CC0 appears to have most of the legal text expected in a proper license including clear allowance of modification and distribution, and lack of warranty or liability.

Especially since they're calling it "Open Speech Recording" you would think that they would be more interested in joining forces with the industry.

It's also interesting that the agreement states, "Google may use the clips ... share the clips with others, including the general public, for example, as part of a public dataset to facilitate research" instead of Google will share the data.

Compare that to Mozilla's Common Voice project which has no agreement.

Notwithstanding the original comment about it probably being a better idea to contribute to Common Voice (it is), this comment is off the mark.

- Common Voice does have a TOS. It's here:

https://voice.mozilla.org/en/terms

- You're equivocating on the word "may". You're reading it as declaration of intent—i.e., as a synonym for the word "might" and def 5 as listed on Wiktionary[1]. In reality, it's being used in line with the most common tendency for things like TOS, RFCs, etc, which is as a synonym for the word "can", i.e., to acknowledge what one is allowed to do (in this case Google), as in def 2.

1. https://en.wiktionary.org/wiki/may#Etymology_1

Then why do you have to agree to Google before contributing compared to Mozilla's, where you can contribute right away?

As to "may", you're incorrect. I'm implying that they should have used stronger language, as in "should" or "will", since otherwise the data contributed to this project may not be made public.

To be fair here, Google is much more likely to get sued, it's not surprisingly they want to be more clear on consent to terms.

Mozilla is a large enough of an organization to have the resources and wisdom to not depend on “Most people won’t even notice” when it comes to legal issues.

Oh, I'm sure a lawyer was involved for both, but different lawyers in different situations are going to have different opinions.

While it is probably hard to understand, the other poster's interpretation is legit as it is referencing an acknowledgement of rights. Without such language, you could provide your voice to this and then when they release, you could say "hang on, I never agreed to give you permission to release this to the public, use for your own purposes, etc."

As far as "may" vs "should"/"will", that'd be very unlikely for them to put into the terms: what happens if they decide tomorrow to abandon the project and just leverage a different data set? They'd now be beholden to release something.

It's clear that you're arguing at this point just to argue. There's no reason to even respond.

Frankly, I'm flaggerbasted how Google can even dare to ask users to give them more data these days.

Sadly, Mozilla's effort is somewhat flawed in its current iteration. There are only very few different sentences (tokens) used as prompts, leading to a lot of repetitions of the same spoken utterance instead of greater coverage of possible utterances, and there is near complete overlap in the sentences spoken in the training set and those in the development and test sets, making it nearly impossible to generalize results from performance evaluations to utterances not contained in the training set. See https://discourse.mozilla.org/t/common-voice-v1-corpus-desig...

Some suggestions for improvements were made in a discussion on the Kaldi Github (https://github.com/kaldi-asr/kaldi/issues/2141), but the bottom line is that it's not clear what the exact use-case of the corpus is. It's not command recognition, there's no particular domain, just more or less random English sentences collected from CC0 sources. Use cases should drive corpus collection.

I thought Mozilla started a similar project? Why not focus on that instead? Why not cooperate?

Huh. This is strange. It's asking me to say "My voice is my passport. Authorize me."

(Hint: https://www.youtube.com/watch?v=-zVgWpVXb64)

Really?

That's almost same thing you say to vocally authorise yourself with HMRC in the UK...

It's something like "My voice is my password".

I wouldn't want me saying that phrase released into public...

wish there were more projects like this

Initial Release of Mozilla’s Open Source Speech Recognition Model and Voice Data

https://news.ycombinator.com/item?id=15808124

https://blog.mozilla.org/blog/2017/11/29/announcing-the-init...

Reply on news.ycombinator.com