Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Speech recognition worse with 8.0

This site may earn commission on affiliate links.
Today, I tried pausing for a second or so before speaking after hitting the button and that seemed to work better.

I think that's what drivers need to do with the new version. Before when we were using it "walkie talkie" style(hold while speaking) recording would begin once you lean on the button. With the new style(press then release) one needs to pause before recording starts. Siri was similar to this(press then release) and you needed to pause a bit before speaking or it would mess it up. Siri is now walkie talkie style which seems more responsive and I prefer on my phone(though while driving I like the new 'press then release" in 8.0).
 
  • Like
Reactions: thimel
The release notes description sounded like you need to press a second time to stop recording. Do you mean it can actually automatically detect that you've stopped talking and begin recognition? It's definitely an improvement if that's the case.
I must have missed that in the release notes but I definitely just stop talking and the translation shows up on the IC. Try it.
 
Hmmm, I feel it's gotten much better than ever before. Gotta wait for the beep before starting to talk and I love that I can see on the driver console what words are being registered/recognized.

Perhaps the accent recognition feature just made it work better for me?
 
I have a theory that the mic(s) were sourced from different manufacturers, or other continuous improvement physical changes have created (at least) 2 very different experiences. Or maybe it's the Pano roof, I wish everyone would put their configs in their sigs...

I know that I can barely use the voice commands in 7.1 (and never with pano roof open) as more than 50% of the time the results are laughable at best, so I'm hopeful with 8, based on the odds here.

So, this just might not be something that the software alone can fix, if it doesn't work we should just report it and bring it in if they ask us to.
 
I have a theory that the mic(s) were sourced from different manufacturers, or other continuous improvement physical changes have created (at least) 2 very different experiences. Or maybe it's the Pano roof, I wish everyone would put their configs in their sigs...

I know that I can barely use the voice commands in 7.1 (and never with pano roof open) as more than 50% of the time the results are laughable at best, so I'm hopeful with 8, based on the odds here.

So, this just might not be something that the software alone can fix, if it doesn't work we should just report it and bring it in if they ask us to.
Works great for me.
 
The new one seems to use an online service to completely transcribe the whole contents. The old one seemed to recognize what you said offline against a small subset of key words that could be commands.

I think it's a trade off which way works better. This way can adapt better to accents and improve better with time because it can leverage unlimited cloud compute resources (or, say, a licensing agreement with Nuance or someone else with more speech expertise). The old way guaranteed mapping what you said onto a command.


(As an aside: I had a 2008 Chrysler where if I pressed the button and made mumbling sounds, it'd say "Calling Towing Assistance". To this day, I can't decide whether that was a feature or bug.)
Both systems use an online service for all of it. The car records a sound file, uploads it to google, who send back a text translation which the car then uses. The only part that has changed is how it's activated, and possibly the audio thresholds of the recording. The actual voice processing is still google and always has been. (this is why you've never been able to use voice at all without an internet connection, and why I lost my voice recognition on 7.0 despite no changes to the car (change was on the google end of the API)
 
The release notes description sounded like you need to press a second time to stop recording. Do you mean it can actually automatically detect that you've stopped talking and begin recognition? It's definitely an improvement if that's the case.
The release notes do say you need to push a second time. However, if you don't, it'll figure it out. The problem being, though, that it waits until it thinks you're done talking, and if I'm on the highway the road noise must trick it, so it sits there listening for quite awhile before giving up.

It's been really flaky for me since 8.0. It regularly fails to recognize the first thing I say. I'll have to pause like someone above mentioned, but that's stupid. I press the button, and it beeps when it's ready. If it's not really ready, it should wait longer to beep.
 
  • Informative
Reactions: cynix
Pausing after the beep (I've been doing almost one second, haven't experimented with different delays) it does work much better. It still doesn't seem to be as good as 7.1 was (also my software got updated from 2.36.108 to 2.38.19 so that could affect things too). I don't have pano roof, nor do I speak with an accent so that might be why it was so good for me under 7.1. Like @Gizmotoy said, it it isn't actually ready at the end of the beep, it should beep later.
 
  • Informative
Reactions: Lex
For those having issues, I'd be interested in if it behaves better while you're connected to wifi (or even a hotspot from your phone). Maybe it's a compression issue based on connection speed.

It seems about the same to me (as 7.1) but one thing that is much better is being able to see what's going on. I used to use a voice command and then miss the "command no recognized" (or several other variations) message that would pop up for 1 second.

Now I can clearly see what it thinks it hears, which usually looks like "Navigate to Starbucks let's go guy are you seriously going fifty-three in the left lane".
 
The second press (supposedly) stops Siri from recording you (at least the waveform visual stops) so it's probably the same here ?

Why is press to talk and the press to stop talking an improvement over press while you're talking and release when you're done? Seems like change for the sake of change.
The button is on the steering wheel, and voice controls are for driving, and I've tried to hold the button down while turning a right angle turn. Though it's brushed off as a chuckle in my day it's not that funny really and probably also wears the button down faster. But then again I only try the voice commands in 7.1 a few times a week if that.

Voice commands are supposed to be the safest way to issue commands to the car when you're driving so it should be really easy to activate. And it should be accurate. My old 2007 Chrysler MyGig (OEM Harmon Becker) and the made-in-Germany (Mercedes) Chrysler mic they added as a part later on for aftermarket MyGig upgrades, did a great job. 10 years ago on a Dodge !

Because this is the future, they should also have many many many synonyms, lots and lots of similar words for the commands. As many as they can think of. It only makes sense to me to help personalize the car and make it easier and safer to control while driving.

Voice commands are also just plain futuristic so IMHO the mic(s) and the voice recognition seem to both still need major attention.
 
Last edited: