Figure 5: High Quality Headset To improve speech recognition I bought a pair of high quality noise cancelling headphones from Costco for $89.00. I realized that this was a crucial part for the project and this headset gives me the ability to communicate through USB wirelessly with 3 feet of range.
We need a 12volt battery to power the motors on the base. The battery used by last year’s group is missing. The 12 volt battery shown below was purchased for $30.99.
Lastly, we purchased another NXT Brick for testing purposes. We did not want to erase the functions on the NXT written from the previous group. In order to continue to add to the software files on the NXT we would have to download new firmware and we might erase their code. The NXT brick is $135.00.
Speech recognition is a complicated task because there are large sources of variability that are associated with voice commands. One type of variability is found in a person’s voice. Acoustic variability’s in ones voice can change the way the computer interprets phonemes. Phonemes are the smallest units of sound and represent 500,000 words in the English language. When the user speaks into the microphone, the user’s voice is converted into digital bytes of information which are comprised of phonemes. The phonemes are then compared to a large library, and when a match occurs the word is then given back to the user. The delivery of speech into the microphone is very critical for this reason.
Another variability associated with voice commands is the user’s environment. External noise associated to an environment can introduce all sorts of problems for the user. For example, a speech program can be fully functional in a conference room where there is no noise, but can be fully not functional when demonstrated in front of a room of noisy kids. For this reason, it’s very important to have a high quality noise cancelling microphone.
Next, it’s important that a speech recognition program is tested in all sorts of environments. This is another crucial step for the reasons stated in the previous paragraph. You can record the accuracy of each command in each environment and predict how successful your program will work when it is demonstrated.
Lastly, I will take two approaches to speak commands to the Bohr Robot. The first approach is to communicate to the Bohr Robot over a serial wired cable and give the robot commands over a wireless noise cancelling headset. Secondly, I would like to add Bluetooth to this robot and control the robot by first giving it a command into the headset and then transmitting that command over Bluetooth.