The Speakularity, Where Everything You Say Is Transcribed and Searchable | Hacker News Reader

The Speakularity, Where Everything You Say Is Transcribed and Searchable

nautil.us

49 pointsdnetesn17 comments

Perhaps its because I grow up in a different culture, but I see completely different use cases. There are already way too much though crimes, and mob lynching over something simple as twitter.

Imagine local lynching club which assembles every Thursday evening (something like bowling), 'data mines' nearest victim and have their way. All with silent support from authorities, because their victim said something nine years ago, which can be quoted out of context.

And once government and HR departments get their hands on this...

Short story "The Truth of Fact, the Truth of Feeling" by Ted Chiang explores this topic:

https://subterraneanpress.com/magazine/fall_2013/the_truth_o...

Hilarious alternatives to the Singularity: https://docs.google.com/presentation/d/1B75jindDAWsm8lBHPl4y...

The notion that we could turn speech in to searchable text in a useful way betrays the author's knowledge of spoken communication. Very often it's not the words that are the important bit. Unless the recording medium can add in 'Person A shrugged' or 'Person B replied 'Yes' but was being sarcastic', any system will be next to useless.

The spoken word would simply become subject to the same channel constraint that the written word already faces today. A huge amount of conflict and misunderstanding in online discussion stems directly from the fact that sarcasm is very difficult to convey reliably through plain text. (A sarcasm tag would work, of course, but half the point of sarcasm is that the listener/reader "gets it". Indicating that you are about to say/write something sarcastic destroys the in-group feeling that comes from being able to recognize and pick up on it.)

Chances are you could already get a sentiment vector with every sound-bite with respect to tone. Body language will clearly need more than a microphone as sensor to do it well, but as RyanMcGreal's comment hints at, the fact that a system like this isn't particularly good is no guarantee that it won't be used.

It'll still be useful, but for worse purposes. To be able to quote what someone says with out being able to tell what is sarcasm will allow you to create great attacks against that person which a significant portion of the population will buy into. Imagine if we had every sarcastic quip from Snowden. I'm sure there is at least one thing he has ever said that would make a lot of people made if it were taken out of context.

Presumably the searchable text is just metadata overlay on the full unredacted HD video stream with high fidelity audio using microphone arrays which allow post-process beamforming and noise reduction.

Even if it's "only" 10-20 years out, we're still all fucked.

English centric view to say the least. If the "speakularity" comes, given the current state of transcription and voice recognition software it is very probable that only english speaking people would be "affected".

French speaker agreeing here, the vocal recognition is currently at least 10 years behind English and I can't get any device to recognize anything meaningful of what I say, I guess for the "speakularity", we can wait !

We're working on exactly that and you can play with it here: http://clarify.io/try-it-now/

While the words themselves are important, it's the context that is even more so. The difference between "that's a great idea" and "that's a great idea" is huge.

We're already working on tone, volume, rate of speaking, and similar but there's plenty of work to do. If you're interested in working in this space in Austin, TX, drop me a note. :)

Reading the first paragraph immediately reminds me of the The Quantum Thief by Hannu Rajaniemi.

In it, there's a colony where every conceivable surface is covered with sensors that monitor every aspect of life. It's stored encrypted in a publicly available database that anyone can query (if you have the requisite encryption key that is).

Very interesting to consider that such a database might not be that far off into the future.

I've been recording everything I say and hear for the past 5 years, with the idea that eventually I'd be able to transcribe it and run all sorts of amazing analysis and insights on the data. Thought, learning and vocabulary evolution. AI replicas. Digital immortality.

Everyone is walking around with a microphone and camera in their pockets. Recording video of people is still too visibly creepy (until we have cameras hidden in RayBan-style sunnies), but recording audio from your pocket is something you can do without causing a shitstorm around you (*yes, legally you should tell people you're recording them).

So we launched a startup a few years ago called Lifebox: http://lifeboxapp.com, but even now it's still too ahead of its time. Obtaining funding was impossible. Once we move past this era of privacy concerns into a sousveillance culture, it's inevitable that everyone will want to record and stream their lives 24/7.

As far as I'm concerned this has already happened.

https://news.ycombinator.com/item?id=9525340

https://news.ycombinator.com/item?id=9492551

I'm looking forward to great speech-to-text technology. I help run fan-driven transcripts for a podcast, and I recently tried using Google's; it returned something like the hosts were concussed, or Markov models of themselves.

Edit: the word salad result: http://mefiwiki.com/wiki/Podcast_107_Transcript_%28automated...

That's partially due to Google's approach. It appears that they're building a single model for every acoustic model, language, and domain. While that works as a general purpose tool, it's less useful as you go down the rabbit hole.

Side question.. how do you host your podcast? I'd love to chat if you get the chance. My email is in my profile.

TLDR: A collection of opinions about the future, based on the assumption that we will record all our spoken words. The piece seems to be inspired by particular Black Mirror episode and offers less frightening perspective.

EDIT: To make my comment sound less harsh, while providing useful summary I made more detailed TLDRio: http://tldr.io/tldrs/55e826a5919e6e1a25000cb2/what-searchabl...

Reply on news.ycombinator.com