Home
Introduction
Construction
Behaviour
Sound Sampling
MIDI handling
Limitations
Sound Toolkit
|
Limitations Due to technical considerations and time limitations, our robot has a few limitations on its performance. Microphone sensitivity Our specification called for the robot to navigate around a room, moving towards areas of loud sound where people probably gather. Unfortunately, the microphones we used are not very sensitive at all, and have an effective listening range of less than a meter. This means that there is no difference in signal between background levels and a person talking more than a meter away, making the planned behavior impossible. Sound location code is still in the robot, and we optimized it to work with very close sounds. A person clapping to the side of the robot will cause the robot to turn to the direction of the sound. Handyboard sampling rate In order to determine the differences between spoken words with a high degree of accuracy, a high sampling rate is necessary for recording the sounds. With the Handyboard only capable of sampling at 500Hz, a high level of accuracy is harder to achieve. As a result, in order to get good performance, a person has to speak somewhat slowly to the robot for commands to be interpreted correctly. The inability to distinguish between many sounds also limits the possible vocabulary understood, as words have to be sufficiently different from each other. Inconsistent sensor behavior At different times of operation, the microphones seemed to behave differently. They gave different readings for identical inputs from day to day, and sometimes fluctuated greatly in readings during operation. Making the robot self-calibrate every time it was switched on, and just before recording commands solved the first problem. This means that a new base level is set, and sound samples recorded can be accurately analyzed. Also fixed by this is the background noise reading. Since this is constantly monitored, the robot does not respond to minor fluctuations, and does not ignore commands if the sensitivity suddenly dropped. The second problem, of fluctuations in readings during operation, proved to be a problem for the sound location behavior. Either of the two microphones can behave erratically, and if the level of one microphone jumps for a moment, the robot will bear slightly to the side that microphone is attached to. Complexity of problem Sound analysis for the purpose of speech recognition is a large and complicated field. There are many algorithms and methods listed on the Internet, many of which require a lot of background knowledge and computing power to program and run. To avoid tackling a problem too big to solve within the scope of this project, we decided to concentrate on the simpler methods of speech recognition. This reduces the possible vocabulary learned by the robot. |
| top of the page |