Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

This is cool - I didn't see any detail of the mic array on the site, where did you find it?

Also, do you think it's feasible for Amazon to keep a voice profile on the speakers? I'm thinking if they are going to tout perfect voice recognition they'll have to make it person-specific at some point.



Well, it says: Tucked under Echo's light ring is an array of seven microphones. These sensors use beam-forming technology to hear you from any direction. With enhanced noise cancellation, Echo can hear you ask a question even while it's playing music.

I've been working professionally with digital audio for nearly 20 years now so I know a fair amount about DSP, acoustics and so on. Very basically, you can measure the acoustical properties of a space by playing a sound known as an impulse and recording the response, and then extracting the acoustical information by a mathematical technique known as deconvolution. This is used in various commercial products for allowing you to simulate, say, the reverberant space of Sydney Opera Hall on a recording made in a vocal booth, or reproduce the signature tone of a hideously expensive guitar amplifier in a cheap DSP-powered device.

When you have hardware where the speaker and microphones exist in a fixed physical configuration relative to each other, as here, then the math gets that much simpler because a lot of your coefficients become fixed quantities. With multiple microphones at fixed distances from each other you can use small discrepancies in the phase of the input audio to infer information about spatial characteristics of the environment. I don't know the exact dimensions of this thing but just eyeballing I'd guess that you could hack this thing to produce a reflectance map with a resolution of maybe under an inch.

Wow, thinking about it I hope it is hackable. Even if you were only able to get the raw input stream from the microphones and had to import the audio to another machine for all the DSP, a perfectly-calibrated speaker + phased microphone array for $200 is a steal.


I certainly hope it's hackable as well. Doesn't look like they're offering any options for developers to get in on it though.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: