Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Jarvis: an Amazon Echo clone in your browser

iceddev.com

Post reply on HN
Print view View original

51–60 of 79 posts

Re: Jarvis: an Amazon Echo clone in your browser

#51

Post by GeorgeHahn »

benlower wrote: post #35
IshKebab wrote: post #3

Yeah it's a clone of the trivial parts of the Echo, but not the difficult parts that are necessary to make it great, specifically: * Beamforming microphone array (this is a real clone: http://www.xmos.com/products/microphones ) * Wake-word / hot-word detection ("Ok Google", "Alexa", etc.) * Intent recognition / NLU

$750 for the mic?!? ( http://www.xmos.com/buy?product=20258 )

Development kits are expensive due to low sales volume, no cost optimization, and buyers' price insensitivity. There is a huge amount of value in having a known good implementation on hand when designing hardware or firmware. Additionally, having the development kit means your firmware team can start developing before your hardware arrives.

If you have a good relationship with them, sales reps will often give or lend development kits.

Re: Jarvis: an Amazon Echo clone in your browser

#52

Post by dharma1 »

xaduha wrote: post #47
whoisjuan wrote: post #45

Earlier quoted context omitted.

I think the Microsoft Kinect comes with a Microphone Array.

PS3 Eye too, 4 microphones AFAIK. Those had a big discount on Amazon recently. EDIT: Still are, 5 bucks. http://www.amazon.com/PlayStation-Eye-3/dp/B000VTQ3LU

I've got one knocking about in the drawer. Let me know if you get beamforming working on it. I think there are multi channel audio drivers for Linux/RPi for PS3 Eye

Re: Jarvis: an Amazon Echo clone in your browser

#53

Post by dharma1 »

IshKebab wrote: post #31
tchow wrote: post #21

Earlier quoted context omitted.

wow, that microphone array you linked costs USD$750! Are there any cheaper alternatives?

Not that I've found yet, but the actual hardware on that board costs something like $10-15, so someone could easily make one. The high price is because it's a development board. (I think it's a silly to price development boards highly, but it is very common.)

I hope a chinese manufacturer makes one soon. The teardown for Echo had the parts for a bit more than that, but you could do it cheaper, multichannel ADC and half a dozen MEMS mics. I guess drivers would be the time consuming bit

Re: Jarvis: an Amazon Echo clone in your browser

#54

Post by dharma1 »

tomc1985 wrote: post #44

What is the current state of offline, non-cloud-connected speech-to-text? My phone has a voice processing chip, and it recognizes my speech pretty well, but I still can't figure out if it's completely disconnected from the cloud (despite intentionally not agreeing to the privacy policy) His demo is just a shim for Amazon's API...

Kaldi is pretty good. Not sure if you can run it on a phone but definitely on a single desktop, purely local processing.

Results depend on the trained model, I think the Tedlium one is alright. And of course quality of input signal - far field/noisy much more prone to errors, that's where the mic array on Echo helps a lot.

Here's a relatively easy way to set it up

https://github.com/alumae/kaldi-gstreamer-server

Re: Jarvis: an Amazon Echo clone in your browser

#55

Post by alexforencich »

outworlder wrote: post #38
IshKebab wrote: post #3

Yeah it's a clone of the trivial parts of the Echo, but not the difficult parts that are necessary to make it great, specifically: * Beamforming microphone array (this is a real clone: http://www.xmos.com/products/microphones ) * Wake-word / hot-word detection ("Ok Google", "Alexa", etc.) * Intent recognition / NLU

I thought that "beamforming" only applied to actively emitting signals, so that the waveforms would cancel/reinforce each other, to get the desired "direction". I have no idea how that works for microphones. Google is not very helpful, I get lots of hits for products.

It's a reciprocal process, meaning that it works for transmit as well as receive. In the receive direction, the antennas/microphones sample the incoming waves at various spatial points, and then the cancelling/reinforcing occurs when the received signals are phase shifted or delayed and summed.

Re: Jarvis: an Amazon Echo clone in your browser

#56

Post by alexforencich »

dharma1 wrote: post #53
IshKebab wrote: post #31

Earlier quoted context omitted.

Not that I've found yet, but the actual hardware on that board costs something like $10-15, so someone could easily make one. The high price is because it's a development board. (I think it's a silly to price development boards highly, but it is very common.)

I hope a chinese manufacturer makes one soon. The teardown for Echo had the parts for a bit more than that, but you could do it cheaper, multichannel ADC and half a dozen MEMS mics. I guess drivers would be the time consuming bit

The board actually doesn't appear to have an ADC, the microphones output pulse density encoded digital data wich can be directly received and interpreted by the processor.

Re: Jarvis: an Amazon Echo clone in your browser

#57

Post by colinramsay »

kayoone wrote: post #11
IshKebab wrote: post #3

Yeah it's a clone of the trivial parts of the Echo, but not the difficult parts that are necessary to make it great, specifically: * Beamforming microphone array (this is a real clone: http://www.xmos.com/products/microphones ) * Wake-word / hot-word detection ("Ok Google", "Alexa", etc.) * Intent recognition / NLU

Even if true, not sure why you wrote such a snarky comment. This obviously is a side project of the author and just serves as an inspiration, it's not like he is asking money for it or anything.

They're not commenting on the quality of Jarvis (an excellent project), but on the headline. One of the key features of Echo, and one which is essential for something like this, is a good microphone. Most off the shelf ones are useless for this, and so the Echo provides fantastic value by delivering not only the voice recognition/action platform, but the hardware to make it work well.

I'd challenge someone to make a practical device that leverages Jarvis. Most microphones just aren't up to it.

Re: Jarvis: an Amazon Echo clone in your browser

#58

Post by eva1984 »

kayoone wrote: post #11
IshKebab wrote: post #3

Yeah it's a clone of the trivial parts of the Echo, but not the difficult parts that are necessary to make it great, specifically: * Beamforming microphone array (this is a real clone: http://www.xmos.com/products/microphones ) * Wake-word / hot-word detection ("Ok Google", "Alexa", etc.) * Intent recognition / NLU

Even if true, not sure why you wrote such a snarky comment. This obviously is a side project of the author and just serves as an inspiration, it's not like he is asking money for it or anything.

It is PR to sell one of their products

Re: Jarvis: an Amazon Echo clone in your browser

#59

Post by dumbguy »

AnkhMorporkian wrote: post #28
kayoone wrote: post #11

Earlier quoted context omitted.

Even if true, not sure why you wrote such a snarky comment. This obviously is a side project of the author and just serves as an inspiration, it's not like he is asking money for it or anything.

I think it's probably because the headline called it a clone, which sort of implies 1:1 feature parity. It's neat, but it's not a clone.

Pedantic stupidity.

Re: Jarvis: an Amazon Echo clone in your browser

#60

Post by MawNicker »

dumbguy wrote: post #59
AnkhMorporkian wrote: post #28

Earlier quoted context omitted.

I think it's probably because the headline called it a clone, which sort of implies 1:1 feature parity. It's neat, but it's not a clone.

Pedantic stupidity.

No. Wow. Clarification is not Pedantry.
Post reply on HN