AI 内容摘要
**Silencio 宣布与 ElevenLabs 合作供应语音数据,目标是采集全球 97% 尚未被语音 AI 覆盖的语言,让数十亿人能用母语与 AI 交互。**
- Silencio 去年将数据采集重点扩展至语音 AI,当前全球 7000 种语言中仅不到 3% 被语音 AI 理解。
- 现有覆盖不足以为全球不到 50% 的人口提供母语 AI 交互能力,而每月已有 2.5 亿人通过语音与 ChatGPT 交互,未来将增至数十亿。
- 语音是人类最自然的沟通方式,语音 AI 让技术首次反过来学习人类语言,而非人类学习技术语言。
- Silencio 公开宣布成为 ElevenLabs 的语音数据供应商之一,ElevenLabs 是语音 AI 领域第一的公司,2022 年成立,估值已翻倍至近 220 亿美元,75% 的财富 500 强企业在使用其技术。
- ElevenLabs 目前仅覆盖约 20 种语言,应用场景已延伸至《堡垒之夜》达斯·维达配音等游戏与影视领域。
- Silencio 未来几年的目标是主导采集缺失的 97% 语言多样性,同时抢救正在消失的数百种语言。
**结论:语言数据采集是语音 AI 普惠化的关键瓶颈,Silencio 与 ElevenLabs 的合作值得持续关注,也是参与贡献语音数据的信号。**
转写文本
today and they're always only possible by the continuous work of the team and the community
contributing to this valuable data that we've been gathering over the years.
Now let's kick it off with Silencio itself.
As you guys know, Silencio has expanded its data collection capabilities last year to
include and to mainly focus on voice AI.
And this meant really creating the goal of capturing all the languages of the world.
And as you guys might know, there are 7,000 languages out there and only a fraction of
these languages are understood by voice AI today.
We're really talking less than 3%.
So imagine all the language diversity that is missing and still left out there to be collected.
And voice AI particularly, we find it to be extremely fascinating industry because you
have to imagine that voice is the most natural way that we as humans can communicate with each
other.
Now think about the since the dawn of time, since human civilization emerged, voice has
been our primary way of communicating.
And it's just that in the last couple of decades that with the arrival of technology,
that we had to learn the language of technology.
And that mainly happened through using keyboards and computers or touch screen for phones.
And the nice thing about the eyes is that it actually flips that back to its original origin,
which is that for the first time technology is learning our language.
And this is where voice AI is going to be taking us to.
Now considering the challenge that we want to solve, the 3% of the language diversity
that has been captured so far is really only giving a platform to less than 50% of all
the global population to be able to interact with AI in its mother tongue.
And if you consider that we're going to be interacting with voice AI on such a deep level
in the near future, it's already happening today by the way.
I mean, we're saying 250 million people a month are already interacting just with
a chat to be dealing through voice.
And that's going to climb to the billions.
So you guys have to consider all the ways we're going to interact with technology.
If it's your smart watch or your smartphone or your autonomous vehicle or even an autonomous
or a smart device at your home, it's all going to be through voice first.
And we really want to make sure with Celensio that AI really gets to understand everyone.
So this means for us, then the next couple of years, we want to be the main company responsible
for capturing all of that missing 97% of language diversity and empowering billions of people
that are left out of this voice AI economy to be able to partake in this future that awaits
us.
And at the same time, be able to preserve hundreds of languages for future generations
that unfortunately are dying out as we speak.
Yeah, I think that's a really big challenge for everyone, especially I'm a firm believer
that I think we really should be preserving languages as it really shows the history of
humanity throughout the thousands of years that we have been here.
But let me just touch on that point.
You said that there's a small challenge of giving a voice to AI or giving accessibility
to the half of the world's population in their own language.
And you mentioned earlier at the same time, 11 labs.
I want to ask you, why does that matter specifically for us or for 11 labs and for
the entire population of the population that does not have access to AI in their own language?
Yeah, no, so I mean, you guys are aware now we dropped that information last Friday,
but we're super proud to announce that we are actively partner up with 11 labs as
one of their suppliers for voice data.
And this is extremely exciting for us.
I think it also brings about, you know, a market fit towards the outside and, you know,
there are many partnerships that we have that we're not allowed to disclose for NDA reasons.
But I think it's amazing that they agreed for us sharing this with the community.
I think it also helps you guys to understand what we're creating here and why it matters.
And 11 labs is, it is the number one voice AI company out there.
And I'm quite sure that most of you have heard of them before.
They had an incredible trajectory.
They started just in 2022 and right now they just doubled the evaluation to almost 22 billion
last September.
And there is a reason behind that.
There's a reason behind that because 75% of the Fortune 500 companies are using 11 labs
in some way or the other.
And this is just the start of the voice AI sphere, right?
And here we're just talking about the first 20 languages of the 7,000 that are out there
and still need to be captured.
And for maybe some, you know, gaming into the us out there that they're even, you know,
help create voices like the Darth Vader voice in Fortnite, another games as well.
So this is not only for the interaction with technology in today to day, but imagine every
single, you know, movie you're going to watch in the future or game you're going to play
most likely.
All of that is going to be primarily driven by voice AI.
So you're going to be able to change mid sentence while you're watching a movie between
English or Urdu or Swahili or French, German, whatever language you can imagine.
And all of that is going to fear as natural as we guys, we would have been having a conversation
right now.
And that that's why I mean, this guy is the limit in terms of what we're going to be
able to create here.
And you have to understand that language is so much more diverse in its nuances than we
could actually consider.
And English in itself is still missing, you know, millions of trainings of data, of hours
of training data in order to be perfected.
There are over 160 or 80 dialects just of English itself.
And most of the labs only cover maybe 10 to 20 of these.
So it's not only the expansion into the other languages, but making the current ones even
more proficient.
Absolutely.
And I don't want to take it back because you're really hinted on it before.
But of course, the opportunity is so big.
That means that 11 labs is not the only player in the game, of course.
And I've seen a bunch of comments and questions on discourse and on X as well is whether 11
labs is our only client this quarter or are there some other ones that we haven't
yet mentioned?
So Silenzio officially already has over 10 clients.
And we closed four important clients in the last quarter alone.
11 labs actually was closed already in the quarter before that.
So I think what is really like astonishing for us is that we don't need a million clients,
right?
I mean, there aren't even a thousand clients that would be relevant for us.
There are probably about a hundred clients that are relevant for Silenzio.
So our client acquisition is different than other companies.
And as you might imagine, it takes months to close the right client to build the trust,
to close our relationship, to be able to start commercializing the data to these clients.
But once you win the trust of these clients, this is where things get super interesting.
Because you guys have to imagine the also cross selling opportunity between these relationships.
You know, 11 labs took a year to close literally 12 months.
Now imagine what is possible once you have them on Slack.
Like literally you could be like cross selling to these companies with one Slack message.
And all of a sudden 12 months of work just turned into 12 days to close an ex deal.
And this is where it really becomes super attractive in the business model itself.
Where we always have been trying to communicate to our communities specifically,
that good things take time.
And we've been here for many years.
We're going to stay here many years.
And we found this super exciting product market fit for us,
where we clearly know the USBs that we have.
And it's just a matter of time until we start closing more and more labs,
as long as we keep on doing what we're doing.
And this is where we need, of course, the help of the community.
You have to imagine that Cylensia already today holds the largest voice AI library in the world.
So language data for all the languages coming after the top 30 languages.
And that already puts us in such a unique position that we're definitely going to make sure
to expand as we go into the future.
For sure, I think it was actually so exciting to hear a lot of these clients.
And to connect that, I think I've also heard a lot of our community asking,
since we just mentioned that we have other clients that we can't name yet.
I think one of the questions they also want to ask is,
what does that mean for Cylensia's revenue?
This is something Theo mentioned, I think a month or two ago, during an interview,
but I think they would love to hear from you about this one.
Sure thing.
So, I mean, we have very exciting news that Cylensia in the last quarter alone
doubled its revenue compared to Q2.
So we reach a revenue of 640K for Q3 alone.
And this puts us cumulative already at over 1 million for this year,
so considering the first three quarters.
And this really meant that since Q1,
we managed to increase our quarterly revenue by over 600% to be precise, 640%.
And this is a trend that hopefully we're going to be seeing to continue.
We're actually expecting to close this year with over 2 million of revenue.
So again, double of everything we did in the first three quarters.
Of course, I mean, disclaimer, none of this is financial advice.
And really, a lot of this really depends on the relationships and contracts
we are continuously delivering and closing with labs.
But we're very firm on the decision and trajectory of this company,
of it really being the perfect one.
Because we really feel we go to a point where we need to hyper focus
on making sure we stick to delivering the best data out there.
And yeah, increase our library and keep on closing more clients.
And what keeping all that notes, what happens once the deal closes,
what happens to a recording after someone submits it.
And what is next in the process of recording itself as well
for our community members to participate?
Is there something new that might be coming out as well?
Nothing.
So I mean, there are many ways to contribute in Silencio.
And the main one, of course, being through our voice AI platform.
You can just go on our website,
silencio.network or silencio.com.
It's the same thing.
And you're going to be able to click on training.
That is going to take you on our platform.
And in our platform, there's a super fast onboarding you can do.
Literally, it takes you two minutes.
And you're going to be able to tell us the languages that you speak,
any other relevant information for you to start recording.
And then you're going to find campaigns in the training I section.
And these campaigns today are either single speaker.
So you just talk by yourself.
You get questions asked in your local language or dialect.
And usually you talk about two to five minutes,
and then you jump to the next clip.
Then there is a multi-speaker experience
where we match you with other speakers of your language.
And for some of the languages where we're actively already
creating data sales, we are offering also transcription services.
And what do I mean by transcription?
And this is super important for everybody here to understand.
The data that we collect becomes
sellable the moment it is transcribed.
So this means that for the finished product
to be able to train the I-models,
we need the audio plus the transcription.
Transcription is literally writing down every single word
in that mother tongue that is coming from that audio.
And we are using the same community to transcribe.
And once that final product is there
and of course for every single client,
we need to tailor a couple of text specifications as well.
That's when we, that's a finish,
so to say, hour that we see as a training,
a hour of data for our clients.
And so far it has been so that,
I mean there's two ways to contribute
in terms of the rewards.
You always get raffle tickets for everything you do.
And these raffle tickets are the ones that you use
at the end of each month to participate
in silencios monthly raffles.
And these raffles are extremely important
because SLC is really what is the main powerhouse
behind the collection of this data worldwide.
And on top of that we are paying USCC payments
for some of the campaigns where we see either a quick...