How to correct Assistant when words misunderstood + how get it to simply accommodate natural pauses?

[Android Central Question]

Two issues currently are really turning me (and doubtless millions of other **native English speakers** off Google assistant. (1) I'm British and although my android assistant is set to (UK) voice. I'm having trouble getting it to understand ordinary plain English speech.

For instance I regularly use it to set alarms. One such important alarm I use often at home is to check the coal fire. I say "Set an alarm one hour from now, labelled coal" Naturally I pronounce 'coal' as in 'shoal' or 'foal' — yet no matter how clearly I speak the blooming algorithm parses it wrongly as "call" as in 'fall' or "ball"...and sets a label "call". Now, those simple words clearly sound different. but damned if I can get Google assistant (UK) to recognise that. Is it weighted to London/SE England, where in their particular non-generic. non-standard accent a person making a phoneCALL is often WRONGLY referred to in speech as "coaler" rather than CALLER for Chrisake?

(2) I see from old reports that this next unacknowledged issue has been driving many users nuts for a long time. While Google are busy showing off how clever their Home algorithm is becoming in 'natural'basic conversation — the android assistant (when listening) seems unable to handle normal very brief pauses while the user thinks... and it FREQUENTLY goes running off to search (or action) with an incomplete and incorrect understanding of what the user intended to communicate...and cannot be held back or corrected. If we can only use super simple short rapid phrases then punching it in via keyboard seems easier, no? Or if have to spend time 'rehearsing' what I want to communicate in one rapid flow... why bother?
Please tell me there is a workaround for such basic infuriating interaction failures... Or at least that Google acknowledge such issues and are working toward fixes.

 

[belodion]

Early days yet, and developments are ongoing.

 

[MotoMartin]

Early days yet, and developments are ongoing.

Google Assistant reply to belodion: "Here are some places near you with developments on sewing"

Yes belodion, I think it is safe to assume that a billion customers (aka perpetual beta testers) realise that developments are always ongoing.

As to answering my questions — It seems I'm left to assume then that there is;
no easy workaround that I could employ to avoid such Google Assistant speech interaction failures of the type mentioned;
no trick to keep Google Assistant's attention for longer than a one second or so pause;
no means to correct the Google Assistant algorithm's repeat misunderstandings or have it "learn" corrections.

 

[Rukbat]

Google Assistant is a VERY early attempt at something resembling what we all think AI is going to be. The problem is that we don't actually know what AI is. Is it just a parser that can distinguish between the noun use and the verb use of the same word? Or is it a semi-autonomous computer? Google Assistant is just a kind of intelligent-appearing front end to Google Search. If you learn how to use Search well, you don't need Google Assistant except for things like turning your screen brightness way down.

 

[MotoMartin]

"...attempt at something resembling what we all think AI is going to be."

That hardly sounds like even a casual software spec.
Doubtless you know AI differs in that given adequate example data and the ability to accept and learn from correction it will improve it's task accuracy and results.

It's dawning on me that Google Assistant seems to have been more of a re-branding and marketing exercise based on voice search, voice command and simple 'tricks' easily achieved by familiar interfacing with other software objects include among which is the open protocols now offered by home control devices. It may be an attempt at presenting something roughly resembling what customers imagine AI would be... But a serious attempt at AI it is not - from the outset it seems it did not have the necessary corrective learning functions built in.

My guess is that yes, the Google behemoth is putting huge resources into AI research, but Google Assistant was just a cobbled together meantime box of tricks reaction to marketing advantage that Apple (and Amazon) were perceived to be gaining.

From reading Google's traditional responses (more like deaf ear) to non-trivial user questions — while doubtless quietly logging those very main issues raised by their billion strong pool of beta testers for design advantage... I would guess that Google Assistant is not being actively improved (having insufficient potential) and will be consigned to the bin in due course - in favour of fresh branding, untarnished by early user disappointment... and whatever they -are- actually devoting serious resources to along the lines of human and algorithm interaction that does have rudimentary learning/correction functionality and hence a smidgeon of claim to be a real attempt at AI.

 

[ps doff]

You can't directly fix Assistant's misunderstanding your speech. The only thing you can do is vote with your wallet and not buy more of it. As long as people don't do that, the Assistant will remain just barely good enough to do scripted commercials.

I want to emphasize Home Assistant. There are several "Assistants" from Google - Android, Chrome, and Home are the main made-for-Google's OSs apps. There are two versions of the Home ASSistant, for the hub and speaker devices. The back end server part is fine, the embedded user interface device apps are mediocre and inconsistent. The Home app is in its 4th major UI revision, and while not awful, it changes so much, so often, and without warning or instruction that it's not very useful. It ought to be turned over to the team that has grown the GMail apps.

It's 2022. Google has introduced tne 3rd generation of Google/Nest audio hardware to work with the Home ASSistant. It's no longer "early days". I have a beef with the Home ASSistant; it's not buggy, but half-baked. It still does an equally half-ASSed job with Google search, but now it badly controls more devices.

I'm from the northcentral US with a very slight Chicago accent. I don't mumble. My voice is annoyingly loud and clear. It penetrates crowds. I speak as if I'm lecturing (hard if hearing partner). I have the beep tone enabled for start and end of voice recognition so I know when the Google speaker device is "paying attention".

When I invoke a Routine using the phrase I created, the Home Assistant doesn't understand me ~30% of the time.

I have one (1) controlled light connectd with a "smartswitch". The swicth is named "light".
"Hey Google, turn on the light" yields 4 differnt responses.
1. "I can't do that yet"
2. "There are no dimmable lights"
3. No response, times out after 15 seconds.
4. The light turns on.

I'm a native-born American engineer with 40 years' IT experience. I cannot fathom how any development team intentionally created the first three responses. None of them are helpful. They are equivalent to duh, I'm stupid and lah-dee-dah.

A very limited assistant should have been able to say "I'm sorry, I have a bad connection, (or "I couldn't hear you" or I didn't understand what you said), please try again." on tne first day it was introduced for public use. This is what a human assistant would say if ii didn't understand a direction.

The Home ASSistant still doesn't understand many US English pronuncation(s). It might work reliably in Seattle but not in Chicago. It's very impatient, truncates listening to questions, returns barely relevant answers, and unless you stop it, will drone on with examples of definitions, and quoting from Wikipedia. If it were a foreign exchange student intern, I would not have retained it for more than a month. It does not know when to STF up.

Sometimes "Set a timer for 3 pm" returns "second time set for 3pm" when there are NO previously set timers, and "Turn on the TV" returns "there are no controllable devices".

The Assistant doing more, differnt things badly is not making it smarter or better. But as a creation of Google, I suppose I should have expected that the assistant wouldn't fall far from the Google, the creator of "prettier, not better" im-Material Design.

An unreliable assistant is worst than none at all. I'm going back to using X10 remotes for device power control, an alarm clock, and browser searches for information. If I can't see a clock or out tne window I might ask it for the time or weather, but my phine and Chromebook do those things better.

The Google ***istant lives up to part of its name. It is a jester that amuses by constantly making stupid mistakes. But it cannot be trusted.