Voice Recognition with foreign accent

I don't have an accent and my 2012 Prius v doesn't understand most of my commands. It always gets the numbers correct, but that's it. My 2011 understands all of my speech. 2012 speech recognition has taken zero steps forward, and about 10 steps back.

 

It doesn't work any better with English in my experience.

All a bit odd since it is a Japanese car. I wonder how the voice recognition works in Japan.

 

I think you need to experiment with how to pronounce things for the car. On my Gen2, I have to say "sev degrees" in order for it to understand "seventy degrees". It seems to understand "go home" much better than just "home". My favorite one is "I've got to go" was answered with "displaying gas station icons".

There was a computer program back in about 1965 that copied files from one place to another. One documented command was "Disk to Disk". The author demonstrated to me that it not only understood English, but was actually bilingual. He demonstrated that "Disco a Disco" also worked. The actual logic was to look at the first letter of the first word, skip the next word, and look at the first letter of the third word. It seems that voice recognition is at the same level of development as word recognition was back in 1965...

 

This is Commander Lt. Pavel Checkov
Authorization code Whictor Whictor 23 Access Stereo
Authorization code NOT Recognized!
Authorization code Wh, wvi, vic tor W.
awh, screw it! Where is mine Ipod!

 

I used to work with guys who were working on voice recognition. I didn't do the work myself, but I picked up a lot of what they saw as problems.

Concentrating on how a machine can recognize the word "four". It's made up of 3 parts, "fff", "oh", and "err". If the "fff" and "err" are dismissed as noise, then the clue that remains is an "oh", or the number zero if we are expecting a digit. To find out if the "err" is part of the recognition, you could experiment with the sound of "foe". Or maybe "foe", "err". In English, machine recognition of the difference between "fore", "poor", and "thore" is difficult. They all have a burst of sound at the beginning. Understanding the difference is easier if there is some calibration available from other speech from the same person. But untrained recognition is going to be, well, poor (pun intended).

You might try "thore" and "poor" instead of "four". I think the machine isn't picking up on the "fff" sound, so making it more explosive might help.

 

Have you tried with someone about the same size sit in the passenger seat?

Maybe the mic is picking up echos from the empty seat but it works when someone is sitting there and absorbing the sound?

 

I got really ticked off with the system not recognizing my commands so I started talking with an angry Japanese accent. I half expected the car to just quit running and leave me stranded. LOL

 

You said that it works better when your Prius is stopped. Speech algorithms tend to be sensitive to background noise so try using a Bluetooth headset that has a noise cancelling microphone and see if the performance improves.

JeffD

 

I would think any noise cancelling system built, would be built around the specific characteristics the mic they plan to use in the system. To replace, you would a mic with similar characteristics. Since the noise cancelling system is not adjustable, it might be an interesting experiment, albeit an expensive one!

 

Voice recognition is a pita. There are so many different accents and pronunciations out there in the English language and unless you speak with a Californian accent you haven't a hope.

Example is the guy earlier who said 'four' is made of fff - oh. How about fff -orrr? or fff -rrroh?

I don't want to talk to my car and I don't want it to talk to me! lol