Is it just me or have speech recognition platforms like Siri actually gone backwards in the last few years (roughly coinciding with the AI/ML craze)?
It feels (anecdata) that Siri is doing a bit better on voice to tex when sending messages but much worse on simple commands like “turn off living room lights”.
This never seems to be a popular opinion on hn, but this is the ultimate Rube Goldberg machine: A complex ai built by a billion dollar company, an enormous amount of man hours, only so that switching a light switch because slower, and less deterministic. I'm certainly not suggesting that there are no valid uses for ML/AI, but digital assistants seem to be an enormous waste.
I'm a fan of Google Assistant's usefulness and I generally agree with this. The Assistant craze seems more justified when you consider that companies like Google and Amazon are using it as a way to build datasets. It's a strategic reserve of data to build other things from (i.e. Google Duplex) and it doesn't matter at their scale if there isn't a vision yet, if a competitor is stockpiling transcribable voice data, they need to keep up.
Ignoring the fact that the research isn't only used one for one purpose, a proper assistant is hugely valuable to me.
Simple things like:
"Remind me to X tomorrow at 9am"
"Add Y to the shopping list"
"Remind me to do Z when I get home"
This is the only way I do reminders now and it's great. It's especially useful while driving, or when I'm in bed nearly asleep and remember something, I can just tell Siri without having to get up and use my phone.
And as far as smart home controls, being able to say "Siri, turn all the lights on" or "Siri, turn the heater on" without having to stop what I'm doing and walk around the house flicking switches is really nice.
Assistants provide a very noticeable QoL improvement for me in many aspects. I think that's more than most other products on the market can say. And that's not even touching on the lives saved from not having to use your phone while driving.
Yes but... Siri is (to me) not there yet. Promising but not there.
Siri is unaware of so many things.
A simple command that would actually be useful (but fails totally):
« Siri, remind me to buy milk the next time I go get groceries. »
And even then... I should not have to mention the part about « groceries ». The request should be perfectly understood with a full stop after the word « milk » and a reminder should pop up automatically whenever I am in a grocery (any grocery).
An even better Siri would also be able to a categorize things properly. For example, if later in the day I say « Siri, I will need chicken, butter, salt and cardamom for my next recipe », Siri should automatically add those to the « milk » next time I go to the grocery.
Anyone know why there is no longer a way to turn off all the suggestions on Alexa anymore?
I've used Alexa since it came out and they've added annoying random suggestions over the years that you were able to turn off in various settings menus.
But in the past year or Alexa has been upselling me stuff or randomly telling me about features I don't want to hear about when I'm just trying to turn on a light. And it seems like there's now actually no way to turn this stuff off.
Does anyone know if you can turn off Alexa upselling or is it going to just be a part of having a smart home forever?
I'm still not very impressed by this, as it can be totally accomplished by a pad of paper and a pen.
I'm believe the most valid use of AI/ML is to perform tasks that people either cannot currently do, or cannot easily do. For example, ai-based up-scaling of old video game pre-renders. It's not really feasible for a person to do this well, unless you simply rebuild everything with a team of artists. And you could argue that up-scaling images for a video game is trivial, since all video games are trivial. But, the point is that the task at hand requires the help of a computer, whereas a to-do list, or using a light switch does not.
I gotta ask, can one not just use their memory and remember to write those things down in a few minutes/the morning? I know that kinda defeats the purpose of a notebook. But, like, remembering to write down to get more garlic should not be difficult in any way. Also, if one is so perturbed at forgetting things then some other questions and areas need to be explored. If one is having difficulty remembering things this much, I fear that there are much deeper issues and possibly some quite serious health problems at play.
1. There's nothing wrong with keeping that same notebook by your bed. Also, you might wake your spouse up by talking to Google.
2. Don't get so messy when cooking, also it hardly matters if your temporary notebook gets a bit dirty.
3. Don't multitask when driving. Taking your eyes off the road is the main concern, but testing has shown driver's voice control systems to be distracting to a significant degree. And, voice assistants are at least somewhat similar.
> tasks that people either cannot currently do, or cannot easily do.
A todo entry or flipping lights via assistants also qualify as such tasks, if we broaden the definition of "easy". The flow from having the thought of an idea or a song to making a note or playing the song by just speaking out loud is just so convenient, without having to context switch from whatever one is doing. Controlling a set of IoT devices with custom commands is another good usecase.
Of course, not everyone has a workflow where a digital assistant fits well today. However, I expect that their usefulness will increase exponentially with time. We're surely heading to the sci-fi future where each house will have a personalized digital guardian responding to the wishes of the family, Jarvis style, no?
Speech recognition goes way beyond digital assistants. Plain old transcription is super useful, especially in terms of accessibility. I was voice coding for a while because of RSI and even as immature as the tech is, it saved my ass.
I tried a few. Caster and talon were the best. I’d recommend talon with dragon as it’s speech engine, but apparently talon has a new built in speech engine which people like, so maybe try that engine first to save money.
I'm not a full data point to work from, but I've got neuropathy and nerve damage. Side effects include severe short term memory problems (I can forget some things as quickly as 10 seconds, as frustrating as that is..) and sometimes motor control issues.
For me, speech recognition and assistants, along with software like Todoist, are able to keep me far more functional than I would be otherwise.
It seems that with AI/ML, choosing/defining your problem well is hugely important. Text to speech, even computer generated natural language is a definite enough task that engineers (and machines) have the feedback to make progress.
Siri is brain-damaged. I am constantly disappointed with how stupid she is compared to Alexa and the gap is widening. I try to keep things as simple as possible with her because she always disappoints me. I wish Apple would open up a tiny bit of those 200 billion in cash reserves on improving Siri.
Same with the Audi. I don’t understand why all the car companies aren’t licensing the assistant technology from the big tech companies and using it as the default.
I get Siri via car play, but even car play is sandboxed because it can’t control things like the radio or temperature.
The procurement decisions for the infotainment system are made around 5 years in advance of the vehicle release. The tech is obsolete before the car leaves the dealership.
Because even the "big" tech from "big" companies is still shit. Have a 2020 LG TV with Google assistant built in, and it's a piece of hot garbage.
Example: I frequently switch between display profiles to suit what I need. Saying
"switch to cinema display mode" - works fine.
Saying:
"switch to user display mode" - 100% of the time results in the TV replying "which user would you like to select?".
Like...it's not my fault that the dumb TV has the custom profile named "User".
Google probably spent billions on voice recognition, but it's all worthless, because someone without an ounce of imagination just coded it to react to the words "change" and "user" as the user profile selector, the rest of the sentence be damned.
But back to cars - Google Assistant in Android Auto is equally shit. Try saying "hey google, open spotify", then "hey google, play music". 100% of the time, it switches from Spotify to Google Music. It's insane.
The move to companies naming everything "Brand Name Generic Name" coincided with the move to voice assistants in the worst possible way. "iTunes" is unique but "Apple Music" is just two nouns - and we see this across so many properties.
For the reason it took BMW so long to say yes to Android Auto - Amazon/Google will happily license the tech but want ALL the customer data off the car.
Here's an article from a few years back, from Toyota publicly stating they won't support Android Auto due to privacy concerns[1]
It's not just Amazon/Google, car manufacturers like Hyundai are selling data to Verisk, a data broker, who in-turn sells the individuals driving data to insurance companies and other entities [2]. They get people to "opt-in" by offering people free 'connected services' which has data-sharing buried in the T&C.
I'm sure other manufacturers are also going to do similar things as this allows them to generate additional, incremental revenue from user data about their end users.
Android Auto at least works, and that in turn has voice recognition that works. It's completely "meta" that I have to use voice recognition on my phone to call a number that triggers bluetooth that activates the hands-free and speaker in the car.
I would argue that in-car infotainment systems should be nothing but a dumb terminal into your phone. We already have devices with voice recognition, navigation, multimedia, etc. Why don't we have a way to just use that?
I've been looking for a single DIN car audio device that is nothing more than an amplifier with Bluetooth, but such a thing does not exist.
My phone already does Music/Radio/Maps - all I need is the car head unit to connect to it and playback the audio through the car speakers.
I've even started thinking about some sort of home built version using parts from a cheap Bluetooth speaker system (minus the actual speakers).
BlueBus is nice; but it works with the stock BMW head unit (nicely supports steering wheel controls tho). It was for pre-bt-enabled BMW's but I think it's able to be adapted to all k-bus-based cars, so roughly 1998-2010?
There's plenty of old-school single DIN devices that have bluetooth in them. You'll usually find they have a radio and other junk (e.g. mp3 over usb) but most are fairly simple and it shouldn't be hard to find one that stays in bluetooth mode all the time.
If manufacturers use big tech voice assistance, it'd mean extra subscription for the car owners...really sick and tired of these "subscription" thingy. Microsoft Office used to offer one-time purchase, now it isn't an option. Are we going to have everything subscription eventually?
Seriously - my guess is that they try to perform the speech recognition client side (on the local hardware) and are less agressive on how they collect data for model training.
Unlike Google or some company in China which make thin clients that send everything to a central server where it is much easier to recognize, correct and train.
Got a 2020 volvo. The voice controls are insanely bad. I can 100% reproduce a situation where saying "hey volvo, switch off the seat heating" turns it up to max.
Same in my Toyota. It has explicit instructions which, despite a few honest efforts, I can't seem to follow well enough for the assistant to work with any reliability. It's also painfully slow as it fails to understand, making the tumble into the pit of infotainment despair seem to happen in slow motion. Not sure how it ever hit the market.
I mean I can understand how stupid siri is. I find it so irritating that I always keep it disabled. But the point that anything Apple doing bad is for the lack of spending is hilarious.
In fact I'd say Apple is just keeping spending money instead of shutting it down. When IBM Watson AI sank like turd in market it cut funding and let most of team go. I think Apple doing same might be more sensible.
Same here. I replaced the few Alexa devices in my house with Homepods over the last year and have notice no difference in how well they work because I don't use them for anything particularly complicated, and I don't _want_ to use them for any more than that. The only thing Siri seems to mess up for my usage is not interpreting things I put on my shopping list correctly, but usually in a way that's still recognizable (like "milk this'll").
Doesn’t default to male voice either, you will have to choose. I bet most people keep it what it always was (female in the US, some countries like UK were male voices).
Lernaut & Haupsie went bankrupt in 99 holding the monopoly of every viable dictation platform and the aftermath of a enormous stock fraud that enabled management to hoover up everything at ludicrous valuations eg IBM Via Voice for something like $4BLN cash deal closed in unbelievable time, provided sufficient obfuscation and destruction to lay the sector to rest for two decades.
Microsoft is making a intervention with things purchase the way I see it.
What they have to do is provide a Linux version or at least O365 interop as good with Edge for Linux as Windows.
Nuance support and products are hopeless - my subscription for $120/yr stopped working with ios14 and the app refused to send password reset emails and then we discovered that no online account management existed and cancelled instead of relying on 8/5/300 telephone queuing.
The thing that makes my iPhone worse is that it's hard to use Google Assistant (haven't tried Alexa) with my iPhone. I can't make Google Assistant the default voice interface. I can say "hey siri hey google", but honestly I'm not doing that. Google Assistant is decent, but it's not amazing. For any non-trivial task, I still have to do look things up manually.
And while I'm ranting... the thing I hate most of about all of these services is they don't have a profanity mode. In the privacy of my own home I have AN EFFIN' HUGE POTTY MOUTH. I love to curse. I LOVE IT. The fact that I can't get one of these services to talk to me like a middle school student is extremely disappointing. I don't want to be politically correct when I'm talking to a virtual assistant in a private setting. I want to say "Hey goognizzle, are there any good mother f***ing movies opening near me this weekend?" and get Samuel L. Jackson style "Here's the mother f***ing movies opening on mother f***ing Friday at the 5 mother f***ing theaters closest to mother f***ing you."*
Edit: TIL how to faux curse on HN. To type display "f***ing", I need to type out "f\*\*\*ing".
(gonna be downvoted to death)
Usually Apple die-hard buyers tend to minimize the importance of software/features that are not sported by an i-device or that doesn't work well there. When the 1st iPhone was released, the lack of native apps was a selling point "we use the web standards". When apps were introduced, they were again a selling point (and still are to the present day).
The day Siri will work equal to/better than other competing systems, it will be used as a main reason for which you should buy an Apple device. The day before that, it will still be something that users actually don't want.
EDIT: and to be a bit more clear that I'm serious with this, it actually makes sense. If a feature doesn't work well, you find ways around it. And if you are used to an ecosystem, maybe you don't even know how well another ecosystem works. And finally, if that ecosystem works for you well enough - or that device - you probably really don't care about that missing feature.
I'm an apple buyer. Siri is terrible. Google assistant is pretty impressive. I would switch BUT - google's integration / support for things like work calendars is so terrible, and their response to consumer / user input so bad (ie, ignored despite #1 request for years and years) that I'm like most Apple users.
Apple (and Amazon) get many things right, and actually seem to pay a tiny bit of attention to user needs. So it's really not worth playing with the google stuff because you end up in these weird nonsensical hells - everything is amazing, and then they just drop the ball in a key corner.
For example, my google work calendar is EASY to integrate with Alexa along with my personal calendar. Fantastic, what's my schedule today works great. Google - falls flat on its' face for this, despite being a paying customer of their Gsuite. Their approach is just full of excuses here, and ignores that this desired interaction works well on their COMPETITORS devices but not theirs.
Just one (of many) examples. They have some sort of eventually consistent backend more often for stuff so you also get weird states that you can't delete things, changes take longer to "flow through" etc.
No, I'm saying that Apple buyers are usually satisfied with their purchase for several reasons already. One feature not working as expected is just ignored/not needed.
My guess is that Apple management doesn't see voice assistants as the "next big thing" to differentiate themselves from the competition. I think VR/AR is where they are focusing and Siri is on life support.
I think so too. I am pretty sure that data will show it too. Five years ago the NextGen computing platform that will change the world where smart speakers. Now it is almost a niche product.
In the flip side I use Siri everyday to turn on lights , timers , custom iOS Shortcut commands , Intercom, play music, and to find my phone when I can’t locate it I’m my house.
I was actually under the impression that this is because Apple actually cares about privacy (unlike the other two) and so is not hoovering up everything you say to it and sending it to the cloud, but rather trying to do as much inference as possible on your device.
As a result Apple's maps and ML assistant aren't as good as Google's which harvests way more of people's data.
I'm perfectly fine with this trade off. Lots of things iPhones do are even presented at as the data never leaving your phone at keynotes/press releases, and ads. A bullet point notably missing from Android phones. I deliberately use Apple's admittedly worse maps application whenever possible because I know it's not telling Big Brother about me when I know Google maps is.
Edward Snowden revealed that Apple has been part of the PRISM electronic surveillance program since 2012.
I would agree that Apple is significantly better on user privacy than Google or Facebook but that does not mean that the NSA isn’t sucking up all your data from Apple Maps.
Claiming that Apple isn’t “telling Big Brother about me” is at best naive and at worst dangerous misinformation that could put activists and whistleblowers at risk of having their location data harvested by the US war machine.
A fair argument but I don’t think it’s really fair to compare Signal to Apple Maps because IMHO while there is a possibility that the NSA is able to somehow intercept my Signal messages on-device prior to encryption occurring; it is almost a certainty that the NSA has some form of back door access to Apple’s infrastructure and is able to access location/search data from Apple Maps.
This isn’t even really a criticism of Apple, it’s just the reality of the world that we have created since 9/11. I don’t believe any company that operates at Apple’s scale can keep its infrastructure totally private and secure from abuses of user privacy by the state. Especially if that company has infrastructure or employees in countries like the US/UK/China that are notorious for having extrajudicial/illegal/secret/unaccountable surveillance programs.
Totally agree with you on Google, they are the worst by far on user privacy. And we shouldn’t really expect anything different from a company that derives almost its entire revenue from advertising/data AND has been deeply connected to the intelligence agencies from Day 1.
I’m always surprised when people shit on Siri. I use it all the time for simple commands, and it works pretty reliably.
I’ve never used anything non-Apple nor have I used any non-phone voice assistant, so I’m not in a position to compare, but as long as it calls the people that I want, sets timers and alarms as needed, and routes me to my city of choice, why would I care?
>I’ve never used anything non-Apple nor have I used any non-phone voice assistant
You're coming at it without experience. If you try the google and/or amazon version, you may still find Siri acceptable but I can absolutely guarantee you will not be surprised anymore when people call Siri shitty, because companied to amazon and google, it is.
I write apps for Android and design devices that run on Android for a living. A friend once asked why I have an iPhone, and my response was pretty simple:
"I'll switch to Android when I can rotate my screen animate the rotation instead of blanking out"
They got it immediately.
-
For those not familiar with Android, your app's UI is completely destroyed on every single configuration change. Rotation. Plugging in a keyboard. Dark mode.
Then the app has to redraw it's entire state from memory in the new orientation.
To do that tiny little thing I described above you'd have to design your app to disable all built in configuration handling (so now it's on you to handle swapping out every resource when a language change happens, or dark mode is turned on) then hand animate every element to its new position on every screen. Needless to say, that's not done.
It was probably very convenient when they were designing the G1 with it's 192MB of RAM, but to me it's a thing that encapsulates everything wrong with Android as a platform for me to use daily.
I wonder what percentage of daily Android crashes in the world are a direct result of this tiny decision. Or god forbid, the amount of gnashing of teeth in how to write Android apps in a "clean" way that manages with this...
(And yes, iOS has state restoration too, but it's strictly for returning from the background, so it's not nearly as intrusive)
I didn't remember how android handles screen rotations so I went and checked, I looked on both my phone and my tablet on a plethora of apps:
- YouTube
- GMail
- FireFox
- Plex
- RIF (Reddit app)
- Material (HackerNews app)
- Main home screen
- Sudoku game
All of them animated screen rotation just fine, with no blank screen or glitching. Just a smooth animation of the app screen rotating from portrait to landscape (and back).
It's possible that your assumptions are outdated and - by your own admission - should switch back to Android.
You realize I work on Android devices all day long right?
I have 5 of them sitting in this room!
You might not notice, but I assure you, they clear the screen then fade back in. The exceptions are apps that have to handle configuration changes anyways like games or full screen video.
It's not always a janky thing, phones have gotten fast enough that the screen redraw is easily hidden behind a half rotate followed by a fade in, but the point is that the hacks are even needed in the first place.
Their approach to configuration changes just adds a massive footgun that trips up plenty of developers. The number of high profile apps with semi-permanent bugs like "I got scrolled back to the wrong part of the page when I rotated my phone!" insane.
A future without that is only coming once we get a replacement for the current UI framework in the form of Flutter or Jetpack Compose (both of which handle configuration changes in new ways)
With limited Android experience I can corroborate the screen rotation stupidity. The latest google wisdom is to use the Model View ViewModel pattern, which can help to work around this issue. If you are using a “view model” its a hack to largely avoid this screen rotation BS, but I don’t expect many Android apps are implemented this way. Keeping up with google is like chasing your own tail.
I haven't written Android code in a few years, but IIRC, there's an attribute you can add to your manifest that lets you handle the rotation without getting your UI getting destroyed and recreated. This is relatively easy to do and seems to be commonly done, at least for bigger apps.
Having worked in both Android and iOS development, I can assure you that they both have some ridiculous quirks and confusing APIs, but you eventually just learn to deal with them.
> To do that tiny little thing I described above you'd have to design your app to disable all built in configuration handling (so now it's on you to handle swapping out every resource when a language change happens, or dark mode is turned on) then hand animate every element to its new position on every screen.
That's not done in "bigger apps", it's done in apps that have large areas not rendered with normal UI elements, like games, or camera apps
I've done some iOS work too, and while iOS has its issues it "defaults" to making better apps, hands down.
It's not unlike the user side of these platforms, iOS has a more opinionated "default" than Android
Jetpack is trying to fix that but it's "not that much, extremely late"
-
But again, this is all missing the forest for the tree here, configuration changes are just a tiny part of the general "backend" choices that add up to a more powerful platform in developing for Android... but a less useable platform as a user
This is not much to do with the speech recognition itself, but with the AI behind interpreting the commands.
I'm indeed also a bit surprised that this hasn't progressed. It's still not possible to say things like "Turn on my living room lights and the hallway as well". It still feels very scripted where you have to say things exactly the right way and in bite-sized chunks to make it work. The same with Alexa by the way.
It's insanely more frustrating because there is no standard list of commands - I'm perfectly fine saying things in the way the computer needs (what is a command line after all) but there's no reference listing what it is expecting so you just have to do trial and error to find out what works.
I would say it has progressed compared to 10 years ago. Speech technology in general to. I remember when YouTube auto captions didn’t even hit close to what was being said we now they’re way more usable.
Siri is terrible, but have you tried Google Assistant? I find it very accurate at recognizing my speech and very fast. It feels like a vastly superior product and I definitely find myself using it on my Google devices.
Google Assistant is pretty good. But Google's on-device voice transcription in GBoard (unsure if it's a Pixel exclusive) is borderline magical. The delay between speaking a word and having it appear on the screen is shockingly short.
Oh yeah, the transcription is super good. I'm so sad it is crippled on iOS by requiring some weird app switching to work. Except for that part, it is super faster there too.
I don't know about Siri, but I do believe Google Home has gotten worse. My girlfriend and I mostly use it only for things like turning lights on and off, pausing/resuming the chromecast, setting timers, and asking for the weather.
I've noticed it's been misunderstanding my girlfriend more and more over the past few months (she has a slight accent to her English, but nothing remotely difficult to understand), and lately it's started misunderstanding simple things that I say too. For example if I have Netflix paused and say "hey google, resume", 30% of the time it will give me search results for how to write a résumé instead of unpausing Netflix. I've had the Google Home for a few years now, and that literally has never happened before now. To get it to work reliably, I have to instead say "resume playback" or "resume chromecast".
Google's appliance recently decided to stream Dark Piano on The Choice on Tune In when I say "OK Google, stream NPR.". Those things are so far apart I assume someone took a bribe to redirect customers for certain keywords. Lol. (Not really but it was my first thought.)
Google's online advice is whenever the assistant fails to work just retrain the device. I remain skeptical that's the issue because it's not like my way of speaking has changed. Instead I have just switched to a new way to request what I need since reporting the issue via Google's vaunted customer service process seems likely to fail.
> Is it just me or have speech recognition platforms like Siri actually gone backwards in the last few years…
I've used both Siri and Alexa daily for many years, and used to rag on Siri a lot. In my experience Siri has caught up to Alexa on most fronts, and I find them more or less interchangeable my common use cases (home automation, timers/alarms, music, news/weather summaries, etc.).
That's not saying much since the Alexa bar is quite low. But like many Apple products, I'd characterize Siri's improvements as "slow and steady" for almost a decade now.
It feels (anecdata) that Siri is doing a bit better on voice to tex when sending messages but much worse on simple commands like “turn off living room lights”.