This section lists all blog posts, regardless of topic.
Microsoft speaker identificationJanuary 5, 2009
Something that I'd like to add to the Grace application is speaker identification. There are two benefits:
1. | If you can identify the speaker, you can switch to their recognition profile which improves the accuracy. |
| 2. | The speaker can use words like "I" and "me". ex. "I have an appointment tomorrow at 3:00 PM". |
Until I get something like this working, it would be impossible for Meredith to make use of a kitchen setup, since the recognition accuracy for her voice against my recognition profile would be unusable.
I came upon an interesting page at Microsoft today:
http://research.microsoft.com/en-us/projects/whisperid/default.aspx. It's neat to see that they're working on some of the same things.
The next six monthsDecember 25, 2008
After working on this project for 6-7 months, we've arrived at Christmas '08, and it is nearly time to direct energies elsewhere:
 | Because I've spent so much time working on this project, I've been procrastinating on my goal of reading the Bible in a year, so I will need to devote 5-6 hours a week on that until June. |
| With the hope of biking across Canada this summer, I need to start getting back into shape, and there is lots of planning to be done. |
| Littlehamberg is set to arrive late March / early April! |
| Work for Wilf on various web sites. |
| Hopefully gone all summer biking across the country! |
With all of these things on the horizon, I don't expect to be spending more than a couple of hours a week on the Plato/Grace projects. Before I completely change gears, I would like to get the following up and running:
| 1. | The plan is to take one of my two LCDs and set it up in the corner of our kitchen with the Voice Tracker array microphone in front and below it. This will require purchasing a used computer (hopefully for cheap) off of Kajiji and putting it below the kitchen counter. Connecting the monitor/mic to the computer will unfortunately involve drilling a 2" hole through the kitchen counter. |
| 2. | Grace will require some changes to reduce the number of false positives: That is, when she responds to a perceived command that is just two people talking in the background. To help prevent this, I'm going to change the grammar to have "Grace" be a command that must be recognized on its own before the speaker can speak the command. Currently the word "Grace" can simply be spoken as the first word of a sentence, and I expect that plays into the number of false/positives. I'll also need to add a "Confirm/Cancel" mechanism to certain commands to give additional control and further reduce undesired actions such as spontaneously playing songs. |
| 3. | I need to finish up exercise 34, which allows appointments to be added. This will serve as the base use case of the system, allowing me to add events and ask queries such as "What appointments do I have this week?" or "What is my next appointment?". I will use this over the next number of months as a test bed for the technology. |
| 4. | Program state needs to be saved on a regular basis and whenever the program closes. |
| 5. | Some basic power management needs to be added so that the computer monitor is turned off soon after a command/question has been answered. This will save energy and make the system more elegant. (Who wants an LCD monitor shining from the corner of the room all the time?) Power management to put the computer into hibernation mode also needs to be added so that when we leave for work and at bed time, the system will automatically turn off to save energy. |
| 6. | An interface needs to be added to the outside world so that, via the web, commands can be given and queries answered from any computer or handheld device. I'll also add a command line utility to my Linux environment at work so that I can type something such as: |
| grace "I have a work appointment tomorrow at 9:30 am" |
|
| 7. | I would like to create a simple, concise web site to present the ideas I've cultivated while working on this project. This will likely be a work in progress over the next few months. |
And with that, it will be time to wrap things up until next fall at the earliest!
Review of the Voice Tracker array microphoneDecember 25, 2008
A few weeks ago I decided to go ahead and order the
Voice Tracker array microphone. It arrived in the mail, and it was one of those times like Christmas where you tear open the packaging and behold something shiny and new.
After doing some testing I can say that I'm very impressed with the technology. Acoustic Magic has done some very good work on this product, and I'm happy with my purchase. (My only regret is spending $50 more for the USB version: I was under the impression that it would mean a better analog-to-digital conversion, but in reality it's just a cheap dongle that most likely isn't any better than your sound card)
The first test I did was using the USB connector, which I figured would have the best quality. The results weren't very good: Only 75% accuracy with a command and control grammar from 12 feet away in the kitchen.
The next test I did was with the standard microphone Jack set at about a 15% input level. I arrived at this setting by using sound recorder and testing what the audio coming in at 12 feet sounded like, which even at 15% I felt was louder than the input that was coming in through the USB connector. (Even with the USB connector's volume set at 100%)
The results of the test was 92% accuracy, which is pretty good considering the distance.
The final test I did was setting the microphone input to about 30%. This resulted in only 1 error out of 40, for an accuracy of 97% from 12 feet, which I was extremely pleased with.
There were still some frustrations, such as it interpreting my saying "17" as "70" more often than not, but overall I can't complain. The other thing to realize is that I'm using XP's voice recognition, which is circa 2001, AKA SAPI 5.1. Perhaps if I were to repeat the test with SAPI 5.3 in Vista I would have more luck with the 17 VS 70 problems. My hope is that Windows 7 will include another update to SAPI, but only time will tell.
Anyway, I am extremely pleased with the results. Thank you Acoustic Magic!
older >>
