Hacker News Reader

Comment by altero | Hacker News Reader

I wish futurist would just drop speech recognition as holly grail. Speech has lot of flaws, is horribly unprecise and non private. I think neural interface has better future.

View on HN

Its worth pointing out that they had to use speech in the film because of the nature of film as a medium. Two hours of people poking screens or "thinking" into a neural interface wouldn't have been very engaging.

They didn't have to: a neural interface could be filmed exactly like speech, just without lips moving.

That's still 2x the work to shoot. You have to do every scene with the actors not speaking at the appropriate times, then you have to bring them all back in to the sound studio to do the neural interface scenes.

That eats into shooting days a lot, and costs a lot potentially.

The poor bastards would have to act through what appears to be a romcom without opening their mouths as well.

Wouldn't it be vastly easier from the sound production angle though - no need to record voices on set for a large part of the production. You'd also be able to use images without worrying about the sound matching, you could push script changes after photography without concern for lip-sync.

I thought they already did that - that's what ADR is, right? https://en.wikipedia.org/wiki/Dubbing_%28filmmaking%29#Metho...

It so happens that recording dialog is my job in the film world. I'd record that stuff ahead of time and play it back on set. It would actually be a lot faster to shoot (but maybe not as much fun to watch).

I actually thought that the speech recognition was a huge plus as depicted in the movie. Since accuracy was no longer an issue there were a bunch of benefits:

- The AI character Samantha could judge mood from the user's tone and volume. There's something intimate and human about vocalizing one's thoughts and it would be great to build on that instead of getting rid of it. - There is some semblance of separation and privacy between computer/AI and user. I wonder how this would work with a neural interface.

The downside is the privacy with respect to people nearby. It's much harder to compose an email in private on public transport when you have to dictate it.

Could we monitor subvocalization? Like voice, but mostly private.

It depends on what you're going for. Relatability seems to have been a major goal of the people who designed the AIs in "Her"; telepathy not being a particularly notable human trait, I think it's reasonable to understand why they might choose a vocal instead of a neural interface. (Especially since the movie is intended to make a point about today, when there is no such thing as a neural interface, but people talk to Siri or Google Now all the time.)

speech is also slow. Very, very slow

Speech is actually pretty fast, but not very accurate. Unless you know stenograpy, it's the fastest way for you to communicate.

Only for people who are comfortable speaking, have no speech impediments, communicate linearly (no backtracking), and are communicating things that resemble natural language. Good for grandma writing an email, bad for a lot of other UX cases...

Reply on news.ycombinator.com