voice anything will help for on the road!!!

Alex, it is my understanding that the Voice Dialer processes the speech locally, while the voice recognition API processes via the same recognition engine as Google 411, over the web. I’m not positive but we might try running it without a data connection.

Good to know. Context must be important. Hopefully they’ll leave it fairly open for developers to tune it the way they want. The yes/no scheme is naturally much easier to implement but their ambitions were high.

And even if shorter words are more difficult, I don’t think anyone would have a problem with saying, “Voice mail” versus “Speaker phone” or some such.

Clap on! Clap off!

Thanks for the link, Tim. I think this will take some more investigation. My sense was that it was processed locally but I could be wrong. Whatever the algorithm is, it needs to be optimized somehow because speech recognition is just not good enough yet, especially with all the conditions a phone has to go through. Background noise, distance for the speaker etc. Not to mention all the variations on people’s voices. BUT if we could get simple recognition amongst 10 to 20 words at a time… wow. Add in adaptation for a particular user… Could be very cool.

I had started a Voice-Controlled SMS app a while back using the unsupported. Voice recognition intent and found it to work well for longer words/phrases but was very unreliable on short words and phrases such as yes and no commands. Maybe had I used the language model option (I didn’t know it existed) that would have been different. I hope it works better with cupcake. I have downloaded but not yet worked with it yet.

The odd voice search results is because the speech recognition uses a google search to get your results, and fine tunes the results according to google searches, not by what the recognition engine actually picked up. At least, that was my interpretation of the code when I added voice for launching stuff in my application.

Look for LANGUAGE_MODEL here:
http://android.git.kernel.org/?p=platform/frameworks/base.git;a=blob_plain;f=core/java/android/speech/RecognizerIntent.java;hb=HEAD

I have a dash mount for my phone. It would be awesome to be able to input directions to Google Maps or control the Music application or TuneWiki with my voice.

In the meantime, I’ll settle for an on-screen predictive soft keyboard so I don’t have to take my phone out of the mount to type in addresses while I’m pulled over.