10

I'm currently working on a smart home with voice recognition and I'm looking for a microphone that I could use for a whole room. What kind of microphone should I be looking for? What does the Amazon Echo and the Google Home thing use?

The microphone will most likely be connected to an Android device. The Android device is connected to an isolated private network hosted by a Raspberry Pi and uses Google's speech recognition in offline mode. Using Tasker+AutoVoice the recognized commands will then call certain http routes on the Raspberry Pi to do stuff with the smart home, like turning the lights on etc.

Would it make sense for me to buy an Echo? Could I use its microphone without an Internet connection?

10

The Google Home and Amazon Echo use microphone arrays to enhance 'far-field recognition' (i.e. recognising your voice from a reasonable distance with good accuracy).

The Echo uses a 7-microphone array (image from iFixit, with the microphones in green) and the Google Home uses a 2-mic array (iFixit; in yellow).

Amazon's 7-Mic Array is open for developers with a commercial use case, although this probably isn't helpful to you, as you aren't buying in bulk for a commercial device.

There was a recent Kickstarter for a product called ReSpeaker which now offers far-field arrays in various forms:

I've not had any personal experience with this product, but it might be a valid option. There are also some ideas on Reddit, such as using the microphone array from a Kinect.

Admittedly none of these options seem particularly friendly to an Android device. It would seem easier to connect the microphone array directly to a Pi or another compatible board, if you find an array that supports it, rather than trying to connect to an Android device.

Regardless, I think searching for far-field microphone arrays should point you in the right direction if none of the above options are suitable.

7

Aurora0001's great answer got me doing some more research and I found some really good information on a lot of mic arrays, including benchmarks.

medium.com did some awesome benchmarking on these. For instance take a look at these graphs that describe hotword detection success rates at different distances from 1-5 meters.

medium.com benchmark graphs

I'm not gonna spoil the whole article, if you want more details, refer to the article here.

But the secret winner, in my opinion, is the PlayStation 3 Eye. It's only $6.99 on Amazon and even cheaper on ebay. And as seen in the graphs, it really does an amazing job. It is USB though and can't do all the advanced fine-tuning stuff and doesn't have an open source firmware. But it is plug&play on the Raspberry Pi. For now I'll definitely get one of those.

  • 1
    That's a really great article you found. Seems very surprising that the PS3 Eye works better than some of the dedicated far-field arrays; be sure to edit this with an update if/when you try it to let us know how it performs for you—I'd be interested to know if it works in a 'real-world' situation. – Aurora0001 Oct 15 '17 at 13:58
  • I was similarly surprised that PS Eye performed so good against its competitors. I did a test at home with an external 3.5mm jack mic, USB headset from Jabra and PS Eye. Guess who won? If only camera was HD though. – antimirov Jul 17 '19 at 7:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.