Live data from Hacker News

Voice control engine that runs fully in the browser using WebAssembly

picovoice.ai

11–20 of 30 posts

Re: Voice control engine that runs fully in the browser using WebAssembly

#12
Coincidentally I tried the python client just earlier today and was blown away. I never even had to record my voice, I just typed two words for the wake word and it was 95% accurate in picking them up even with heavy background noise.

I am definitely going to integrate the wake word functionality into my Chrome extension that lets users control the browser via voice (https://www.lipsurf.com) so they can turn on the extension without even hitting the button.

I thought I would need to look into compiling into web assembly myself, super impressed that this too has already been done.

Re: Voice control engine that runs fully in the browser using WebAssembly

#13
post #9

If you are interested in a voice engine which will be open-sourced soon and works on Raspberry Pi 3, iOS, Android, osX and Linux, you can take a look at what we are building at Snips https://snips.ai

> will be open-sourced soon

Great! When that happens, I'll consider it.

Re: Voice control engine that runs fully in the browser using WebAssembly

#14
post #5

Each of the >30 times I've hit F5 on this site in Chromium it causes a 100% reproducible, complete tab crash. The tab fully loads and appears properly for about... 100ms or so beforehand, so I'm pretty sure it's to do with some kind of wasm interaction. I have my microphone set to "ask before accessing" so I don't think the crash is to do with audio. FWIW, I've been experiencing periodic tab crashes for the past few…

Why are you not up to date? Chrom/ium has a stable extension API; it's not like it should break anything, and you're missing security and bug fixes.

Re: Voice control engine that runs fully in the browser using WebAssembly

#15
Looks like this is source for the wake word engine: https://github.com/Picovoice/Porcupine

But their site says to email for licensing, which bodes poorly.

EDIT: License for "Porcupine" is Apache but with this:

THIS LICENSE APPLIES TO ASSETS IN THIS REPOSITORY INCLUDING DOCUMENTATION, SOURCE CODE, BINARIES, AND RESOURCE FILES EXCEPT FOR THE OPTIMIZER TOOL, ITS ASSOCIATED RESOURCE FILES, AND KEYWORD FILES GENERATED USING IT COLLECTIVELY LOCATED UNDER Porcupine/tools/optimizer. THE OPTIMIZER TOOL AND KEYWORDS FILES GENERATED USING IT ARE NOT GOVERNED BY THIS LICENSE. SPECIFICALLY THEIR USE IN COMMERCIAL PRODUCTS WITHOUT ACQUIRING A LICENSING AGREEMENT FROM PICOVOICE IS STRICTLY PROHIBITED.

Re: Voice control engine that runs fully in the browser using WebAssembly

#16
post #8

Oddly, all site text invisible in Chrome 55 (UC Browser) and FF 57. (text exists, but must be same as bg color, including all buttons)

It's ok with FF 62 on Android, but it'sshouldbeok with older versions. Being on a tablet I can't check if they're using something new.

Re: Voice control engine that runs fully in the browser using WebAssembly

#17
post #6

Interesting tech, but on a very restricted demo it was wrong a huge number of times for me. I'd recommend adding a note that the lamp itself will go red in response to "ok lamp", at least I think that's why it was doing that - it was a bit confusing and to see the first thing happen after saying "ok lamp, blue" having the picture of a lamp go red was not hugely encouraging initially. Perhaps add "listening for colour…

Weird... I just tried it and it worked perfectly for me, even when whispering away from my phone's microphone. And I have a strong accent.

Re: Voice control engine that runs fully in the browser using WebAssembly

#18
post #6

Interesting tech, but on a very restricted demo it was wrong a huge number of times for me. I'd recommend adding a note that the lamp itself will go red in response to "ok lamp", at least I think that's why it was doing that - it was a bit confusing and to see the first thing happen after saying "ok lamp, blue" having the picture of a lamp go red was not hugely encouraging initially. Perhaps add "listening for colour…

Sounds like a good idea. I will think about it. thanks for feedback.

Re: Voice control engine that runs fully in the browser using WebAssembly

#19

Just to add my experience, the demo worked perfectly straight away (Chrome 70 on macOS). Really cool to see this tech, hope it's easy to integrate.

thanks a lot for the feedback :)

Take a look at this standalone demo if you like https://github.com/Picovoice/Porcupine/tree/master/demo/js

it should be a matter of tweaking a few parameters.

Post reply on HN