News for hackers. Stuff that matters.

A comment on The Native Client Proving Ground (Chrome). Parent comment

Re:The Native Client Proving Ground (Chrome) by denniskane on Sunday November 15, 2015 @10:19AM (#10569799)
Emscripten running a crappy voice synthesizer might work, but nothing close to the silky smooth voices that chrome includes. It will really just add tons of bloat to the page. Voice synth is truly something that needs to be baked into the browser.

Are you seriously implying that pocketsphinx in JS will do anything close to the level of voice recognition necessary to recognize pretty much anyone's voice, sight unseen? I mean, really? You kinda need to train the hell out of that thing to get it to recognize the difference between "merry christmas" and "happy hanukah", and even then, it will only really work for one person.

[ Parent | on HN ]
Re:The Native Client Proving Ground (Chrome) by pcwalton on Sunday November 15, 2015 @12:38PM (#10570297)
> Emscripten running a crappy voice synthesizer might work, but nothing close to the silky smooth voices that chrome includes.

> You kinda need to train the hell out of that thing to get it to recognize the difference between "merry christmas" and "happy hanukah", and even then, it will only really work for one person.

So compile an engine you like better.

You claimed that voice recognition and synthesis "could not possibly be done in any other browser". As in, it's impossible. The fact is that you can do it by just compiling your favorite engine to asm.js, and people have done just that with several of them. So your sales pitch for Chrome is wrong.

If you want to say that it's easier in Chrome because of the Web Speech API, that's fine. But impossible in other browsers? I don't believe it.

[ Parent | on HN ]
Re:The Native Client Proving Ground (Chrome) by denniskane on Sunday November 15, 2015 @12:52PM (#10570339)
I don't need to sell anything. I've already proven my point by building something that actually works. If you want to prove something about your favorite technology, you are going to have to build something that actually works using it.

Given the current state of browser technology, the simple prototype that I've created could not possibly have been done with a dumb little JS thread running in a dumb little Web Worker.

What in the world does it mean to "compile an engine I like better"? Do you really have any clue how much sophistication is going on server side with the SpeechRecognition API? Am I supposed to just magically conjure up something that can be compiled in order to recognize arbitrary words from arbitrary people?

[ Parent | on HN ]