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

Voice/speech recognition sensitivity and accuracy

This site may earn commission on affiliate links.
I work for Nuance, who provides voice recognition technology for most voice-enabled automobiles (as well as mobile phones, TV's, Dragon Dictate, etc.). I also just got my Model S this past weekend. Based on my experience, the failures we're seeing are network-related, e.g. Google's servers are not responding in time. Nuance's system uses both embedded and networked recognizers, and uses whichever result it has more confidence in (or falls back to the embedded recognizer if the server doesn't respond promptly). I haven't played with the MS system to determine whether there is an embedded recognizer (can it recognize anything if you're in a location with no network connectivity?), but if not then poor network or server responsiveness could explain what's being seen.

- Bill
 
The system continues to work well for me for English titled songs, although it occasionally will still fail.
I wonder if the servers were extra busy over the weekend.
I do think I experience occasional server failures that lead to an unrecognized command.

It's generally predictable whether the song will be hard to parse.

For instance, "Listen to Panic Attack by Dream Theater" is all English and should be easy to parse. Something like that will almost always work for me. Occasionally, I'll get an "unrecognized command" and there's really no reason for the failure other than server issues.

I've found that pauses between words that aren't as English-y help substantially.
"Listen to Chauffeur by Duran Duran" might fail as an unrecognized command, but "Play ... Chauffeur ... by ... Duran ... Duran" should work fine.

Sometimes, I think the title is just bound to fail.
For instance, can you make
"Listen to Bodhisattva by Steely Dan" work? I tried it once fast, and it was an unrecognized command. Slowed down, I got "Booty Shot Love by Steely Dan". :)

Wishing for classical music is next to worthless, although the results can be entertaining:
"Listen to Samson and Delilah by Saint-Saens" gave me "Samson and Ally love bisexual".

It seems my car has a dirty mind.
 
The voice recognition system definitely works better for some than others. For me the hit rate has always been so low it's not worth bothering, and I'm fairly certain it's not a network thing. It just doesn't like my voice.
 
The system continues to work well for me for English titled songs, although it occasionally will still fail.
I wonder if the servers were extra busy over the weekend.
I do think I experience occasional server failures that lead to an unrecognized command.

It's generally predictable whether the song will be hard to parse.

For instance, "Listen to Panic Attack by Dream Theater" is all English and should be easy to parse. Something like that will almost always work for me. Occasionally, I'll get an "unrecognized command" and there's really no reason for the failure other than server issues.

I've found that pauses between words that aren't as English-y help substantially.
"Listen to Chauffeur by Duran Duran" might fail as an unrecognized command, but "Play ... Chauffeur ... by ... Duran ... Duran" should work fine.

Sometimes, I think the title is just bound to fail.
For instance, can you make
"Listen to Bodhisattva by Steely Dan" work? I tried it once fast, and it was an unrecognized command. Slowed down, I got "Booty Shot Love by Steely Dan". :)

Wishing for classical music is next to worthless, although the results can be entertaining:
"Listen to Samson and Delilah by Saint-Saens" gave me "Samson and Ally love bisexual".

It seems my car has a dirty mind.



I agree that classical music is a total bust with the voice recognition software, but that doesn't matter as slacker can't really do classical when typed using the keyboard.

BTW regarding Sampson and Delilah by SS, there is actually an orgy scene in the opera. I was very happy to see my wife and daughters sleeping soundly in the seats near me.
 
I work for Nuance, who provides voice recognition technology for most voice-enabled automobiles (as well as mobile phones, TV's, Dragon Dictate, etc.). I also just got my Model S this past weekend. Based on my experience, the failures we're seeing are network-related, e.g. Google's servers are not responding in time. Nuance's system uses both embedded and networked recognizers, and uses whichever result it has more confidence in (or falls back to the embedded recognizer if the server doesn't respond promptly). I haven't played with the MS system to determine whether there is an embedded recognizer (can it recognize anything if you're in a location with no network connectivity?), but if not then poor network or server responsiveness could explain what's being seen.

- Bill

Bill, thanks and please please be sure to update any info you gain. Seven year Dragon user at work, 15 pages minimum daily. VR in MS has become worse since 4.4 update - that is my take. Appreciate all the input above - will put to use and hope it is all server-related and therefore subject to a quick fix.
 
Bill, thanks and please please be sure to update any info you gain. Seven year Dragon user at work, 15 pages minimum daily. VR in MS has become worse since 4.4 update - that is my take. Appreciate all the input above - will put to use and hope it is all server-related and therefore subject to a quick fix.

I'm pretty sure it's server as my issues went away. I had issues again yesterday, but also had issues streaming on slacker, so I think it was a network thing.
 
Does the car "learn" your voice over time? I would have assumed not, and it may be just coincidence, but the brand new oaner we had for 2 days last week had a hard time understanding me and it reminded me of the early days with my car. FWIW I'm really happy that Model S seems to have gotten used to my accent and I really like the function esp. on the nav system.
 
I have issues with voice commands, but also have issues with phone calls. i believe both are related to problems with the microphones. i rarely use the voice commands because they never work. when i record a voice command, sometimes after letting go of the button i get played back loud static. does anyone have any experiences similar to what i have?
 
I have issues with voice commands, but also have issues with phone calls. i believe both are related to problems with the microphones. i rarely use the voice commands because they never work. when i record a voice command, sometimes after letting go of the button i get played back loud static. does anyone have any experiences similar to what i have?

This is my situation as well. The voice command thing is more intermittent...but then the phone can be too. We've used 3 different phones (2 iPhone 5s and 1 Droid) and all 3 have times where the outgoing call is all but unintelligible. That's not the server, that's something that's either firmware or microphone based.