I'm enthusiastic about the potential for this technology but I don't like either of these options. I feel like both the customer and the product. I'd like an open source, neutral voice assistant that is working for me, not trying to exploit me.

I stumbled upon kalliope recently (https://github.com/kalliope-project/kalliope). I tried it a few weeks ago and was pleased with it in general but it still needed an online connection in order to parse your commands. According to the changelog they added the option for an offline STT recognition now.

Neither of those options is very scriptable either. They just kinda suck. None of them have any compatibility with actually used home automation ("computer ! lights !" is actually useful). You're right. We need an assistant that's just a piece of software. Paid for, working for me, not caring which messenger it has to use.

Yup. I'm not really going to be interested in this landscape until i can get a full NLP / Intent Recognition / Voice Transcripting package in my home, completely offline.

I'm not even paranoid about the NSA/etc either, i'm paranoid about a potential 3rd party listening in and knowing when to rob my home, or viewing my family through the IP camera, etcetc.

I want a smart home. I don't want an internet connected smart home. It's simply too dangerous.

Scriptability, security, privacy, latency. The list goes on.

Intelligence services of the world, are you listening? There is something to learn from the way these companies have managed to create themselves an army of informers. First they managed to convince people to pay for their own personal tracking devices, putting one in nearly every pocket. They come equipped with microphones and cameras, with satellite positioning, with everything you could wish for to keep tab on your subjects. Next they managed to convince them to use those devices to create a detailed profile of their interests and their friends and relatives, there to be perused at will. And now they've gone a step further, convincing them to put microphones in their domiciles, always listening, net-connected, recording all the time. With voice recognition to ease the transcription of anything of interest. If anything these commercial intelligence services suffer from an overload of data on their subjects, not a dearth of such as is the usual case for governmental services.

Note to current and future governments: outsource intelligence gathering, it is much more effective that way.

Voice is just an input device like a keyboard is.

Why do we see companies fighting over this technology, but didn't we see those fights back in the days when the keyboard was invented?

Also, why is open-source lacking behind? Surely there's plenty of free training data available, so that can't be the reason.

Because this stuff is very compute-intensive and that is simply expensive.

Also, I don't think there is as much free training data publicly available as Google or Amazon have gathered privately.

> Voice is just an input device like a keyboard is.

Also, how would people feel if their keyboard would need to connect to Google or Amazon in order to operate?

Without having any "hands on" (or off) with either of these devices - Can anyone tell me if there is any training data held locally on either of these or do all voice commands get sent to Amazon/Goolge for parsing centrally?

In the case of Echo (I have one), it only sends to data to Amazon when her name is heard. The training for the detection of that name is local to the device, which is why there are only a few name options (Alexa, Computer, Echo and Amazon)

Do either of the devices do actual voice recognition or do they only have speech recognition?