Do you mean Dragon Speech or how it was called?

Anyway, old speech recognition software was quite horrible. Most did not even worked without prior training. And Google does have now offline-speech recognition too. But true, the ability to trigger with any desired phrase is something still missing.

The ability to trigger with any desired phrase is easy, but not done for privacy reasons, to reduce the chance of it accidentally listening to irrelevant conversations.

The inability to change it from Hey google is done for marketing / usability reasons.

Nuance Dragon NaturallySpeaking

What was the software name, if I may ask? I remember speech recognition pre-CNN to be quite terrible.

Microsoft had speech recognition since WinXP. And also Dragon Naturally Speaking. Both needed a couple of hours of training, but worked really well, completely offline, it was amazing for me at the time. It did have a very high processor usage, but that was on freaking single core Athlon and Pentium. I'm not even a native English speaker, though dare I say my English is on par with any American.

You're talking about different concepts.

Voice recognition used by things like Google Assistant, Siri, Cortana, and Alexa usually relies on a "wake word", where it's always listening to you, but only starts processing when it is confident you're talking to it.

Older speech recognition systems were either always listening and processing speech, or only started listening after you pressed a button.

The obvious downside of the older systems is that you can't have them switched on all the time.

I think it would be really easy to create an app that would also listen to a very specific phrase (like "Hey Merlin", simple pattern match, with a few minutes of training for your own voice) and then start Google Assistant.

It's so embarrassing saying Hey Google all the time, and for me, it just feels like I'm a corporate bitch, tbh. It's true, which just makes me feel worse :D