Since the processing is cloud based, and they know my identity, why don't the devices recognize this fact and cooperate. Instead of just 7 beam forming mics in the Echo, if you have two within hearing distance you could have the benefit of 14 and a unified response.
The echo and noise reduction software that I'm aware of can't really do that in a reasonable fashion.
With current solutions, you've got one DSP that's receiving all the audio streams simultaneously, and they need to be exactly synchronized in time. Then, using basically pattern-matching, it figures out what direction the user's voice is coming from, and combines some/all of the audio streams together to eliminate environmental noise and make the speech as clear as possible.
To do this with separate devices, you'd want extremely precise time synchronization. Which is possible, but I wouldn't want to implement it.
The extra processing and synchronization would take longer, and delay input to the speech recognition engine. I don't think it would enhance the user experience.
Just have the Echo that hears the person best be the one that responds. So simple, and easy to implement. I honestly don't understand why Amazon hasn't fixed this yet. It's so fucking obvious.
Ah yes, the rally cry of the person not doing the actual development work... In my experience, rarely is _anything_ "So simple, and easy to implement".
> I don't think it would enhance the user experience.
Baidu trains the voice recognizer by adding all kinds of noise to the training data. I think it might be easier to do that than use multiple microphones. The neural net learns to do the difficult process of separation of useful data from noise.
The echo and noise reduction software that I'm aware of can't really do that in a reasonable fashion.
With current solutions, you've got one DSP that's receiving all the audio streams simultaneously, and they need to be exactly synchronized in time. Then, using basically pattern-matching, it figures out what direction the user's voice is coming from, and combines some/all of the audio streams together to eliminate environmental noise and make the speech as clear as possible.
To do this with separate devices, you'd want extremely precise time synchronization. Which is possible, but I wouldn't want to implement it.
The extra processing and synchronization would take longer, and delay input to the speech recognition engine. I don't think it would enhance the user experience.
Edit: spelling.