View Single Post
  #59  
Old 12-31-2011, 01:35 PM
davesf's Avatar
davesf davesf is offline
2012 550i x-drive DSB
Location: san francisco
 
Join Date: Oct 2011
Posts: 147
Mein Auto: 2012 550xi DSB
Unlike the title of this thread, I don't think the BMW Navigation sucks, I just think it needs some love in a couple of the right places....

My main point is that it would be great if the BMW engineers would integrate readily available and commonly used text-to-speech technology into the BMW nav system for spoken streetnames. It would make the voice nav experience better. A few more details below...

Quote:
Originally Posted by tadtaggert View Post
Things are very different from the 80's. While speech recognition is a more difficult proposition, remember that with respect to the BMW system you're dealing with an extreme subset of words/phrases (basically a command set),
The voice command system both recognizes a subset of commands, and recognizes a large-dictionary of proper street/city/state names.

Speech Recognition for a small subset of commands is much easier than for large dictionaries. Recognizing proper street and city names requires a large dictionary, and thus is relatively hard. Nobody has a nav system that does this well, because it's so hard to do. The technology is called "untrained speaker independent large-vocabularity recognition" and there is no system that works very well. They work abysmally in the face of background noise.

Google Voice Search (on mobile phones), in my experience, is the best "untrained speaker-independent large-vocab" system for place names. It performs so much better because it puts a whole rack of computers to work each time a single-person's voice input arrives (hundreds to thousands of computers). As a result, it can do thousands of times more processing than can be done on a single computer in a car -- and it's still not perfect.

---

Speech Synthesis for arbitrary text, on the other hand, has many acceptable solutions for mostly phonetic languages. Almost all consumer nav systems have this feature. Operating systems such as Windows, MacOS, and Android have text-to-speech built in as a system service.

Looking at these lists of TTS engines, there are at least two completely free TTS implementations (Festival and FreeTTS), many more semi-free, and 6 commercialy licensable versions. Microsoft's TTS engine has amazing pronounciation, and is based continued work by the folks who made the free Sphinx-II engine.

This is an easy thing to integrate into the Nav. I hope they do it sometime soon!

Quote:
What you call 'easy' to implement may meet the standards for Kia or Hasbro, myself I always have turn guidance speech muted anyway, don't need nor want the distraction.
This sounds like some kind of elite brand rationalization. Any street-name TTS would be better than the current 1980's Magellan style "turn in 500 feet". There is nothing "elite" or "quality" about it.

However, it also doesn't sound like you are interested in this spoken-streetnames issue at all, so it's strange you seem to think the rest of us should suffer. I assure you those of us using voice directions would prefer it to speak all streetnames.

Anyone who prefers the "turn left in 500 feet" please speak up now!

----

On a side-note, it would also be awesome if I could use Google Voice Search from the car... either by BMW integrating directly with Google Voice Search, or by BMW/Android supporting hands-free multi-function button use. (At a guess, I think BMW doesn't send bluetooth-multi-function button signals over bluetooth, and Android doesn't yet turn them into voice-search, so both sides may need to be improved). However, AFAIK this isn't something anyone else supports yet either. Maybe it'll be a BMW first.
__________________
2012 550xi Sport - DSB / black nappa / anthracite / ceramic controls / and lots of trimmings...

Last edited by davesf; 12-31-2011 at 01:40 PM.
Reply With Quote