Hacker News Reader

Comment by brtkdotse | Hacker News Reader

Never mind AI, I just want Siri to have better text-to-speech quality than my freaking vacuum cleaner.

It's incredible how bad it is. It actually seems to be getting worse year by year. I nearly crashed my car on the freeway trying to set a reminder to pay a toll the other day - I changed my mind and tried to get siri to cancel the action, and no matter what I said, she kept asking "what time? what time? what time?" on endless loop, and when siri is activated on my car's dashboard, I can't see the map - forcing me to avert my attention from the road, disconnect my phone, then plug it back in and pull my map up again.

She can't do or understand the most astoundingly basic stuff. I guess maybe it "feels" worse now because most LLM's are pretty good at understanding your meaning/intention, but my god, it's so bad that if I were in charge of that product I'd rip it out entirely. There's no way anyone finds any real use out of it.

Even if everything worked you should not be trying to do that while driving. Keep your attention on the road - you are in control of several tons of metal moving at deadly speeds. (don't sing a long with the radio either)

I'm unsure of your point. Are you arguing that issuing a verbal command to my voice assistant is not paying attention to the road? No more than talking to a passenger, really.

Yes. I'm arguing talking to a passenger is not something you should be doing either. You are in control of something deadly, not a video game or toy - act like it.

This seems unnecessarily snarky and way off topic, not to mention a bit silly. I've piloted much, much larger things professionally than a car, thanks for your concern though.

FWIW you were not forced to distract yourself from the road - officially, you should have got off the freeway at the next exit and parked before sorting stuff out.

When something is called CarPlay and actively marketed as a way to use your phone services whilst driving, you would expect it to have a good enough UX that doesnt detract from safe driving to use it.

The law doesn't care what marketing you received and neither does physics. Don't crash your car over a radio.

I didnt, I wont, and nor did the parent post. You seem to have conjured up a scenario just to be condescending. Dont do that. CarPlay or anything else designed to use while driving should never fail to the point of distracting the driver. If it does, its broken and not fit for market.

IMO in this instance, Sirri should have recognised that CarPlay was activated, that it had not got an answer it recognised multiple times in a row, and failed safe saying 'Pull over to continue' or something like that. Hiding the current navigation map and asking the same question over and over again to the point of distraction shows the system is not fit for pupose and shouldnt be used whilst driving.

I mean, sure. Or this could just work how it's marketed to work. Thanks for the helpful reminder!

You could have just given it a time and canceled it later when it was safe to. No need to endanger others on the road for this.

Had I not been flustered and in insane traffic I probably would have been able to figure this out on the spot quickly enough. I did eventually try this after a few minutes and it still wasn’t working. Alternatively, the product could just work how it’s advertised.

I need better speech to text. I have to repeat my text message 5+ times sometimes when sending it through carplay. I'm not talking about anything long either. A handful of words that my accent just doesn't work with. More than once per day I get fed up enough to pull over and text it.

I have a bunch of "premium" smart outlets(eve energy, some are "THREAD" enabled, all the bells and whistles) all named and connected to things I want to toggle with voice: a small radiator, an air purifier, a humidifier, some accent lamps, a fan, an infrared therapy lamp etc. -Before anyone asks, I ordered the European version of their energy strip for higher wattage stuff-. I give each one a clear and unambiguous name, and still, at least one in ten times Siri will be confused about what I'm asking and turn off EVERY SINGLE OUTLET IN THE ROOM. It's endlessly frustrating. For what it's worth the outlets never de-sync or disconnect like the random amazon ones do so I blame it purely on Siri.

Hey now, Siri’s speech-to-text quality is also worse quality than a vacuum cleaner’s too.

siri’s speech recognition and intent handling is so comically bad. as of late, for some unknown reason, when i ask siri to favourite the current song while driving it curtly replies that it doesn’t know which speaker i’m referring to. this used to work.

another fun problem is siri not recognising my speech despite me not having a particularly strong accent, speaking slowly, and enunciating. i’ve gotten into the habit putting on a valley girl or bbc news anchor voice while using siri since that usually works.

whoever is in charge of siri needs a reality check, the feature is borderline unusable.

And the vacuum cleaner really sucks

It’s gotta be just barely holding on in some legacy environment right now. I’m surprised they don’t just start over using new tech. Maybe that’s the end goal with on-device inference to sunset cloud Siri.

I think it’s at least partly cost reduction. Push everything to the phone so people have to pay more for phones and they pay less for infra. Win-win.

Otherwise why don’t they let older devices use their server-side private LLM setup?

OpenAI whisper runs on iOS native on device, can't be just costs. https://github.com/openai/whisper/discussions/407

That’s AI though…

Reply on news.ycombinator.com