Speech-control is one of the most important human-machine interactive methods. Now, most of the research of speech-control is based on PC and without consideration of the noise. But the environment where robot works is various, so the speech signal may be corrupted by various noises. Without eliminates them, the corrupted speech signal maybe useless for us. In this thesis,some effective signal processing methods are adopted and optimized with the application platform-TMS320LF2407. And them can be separated into three parts: 1. Signal Conditioning: A good analog front end is a need for digital signal processing. As for the speech signal, on one hand, it should be amplified because the output signal of microphone is feeble, on the other hand,it is a wide band signal and the sample rate of analog to digital converter (ADC) is finite,so an anti-aliasing filter is need to be added before it gets into the ADC. In this thesis, as a trade-off,a 3rd order Butterworth Low pass filter based on Multiple Feedback (MFB) topology is adopted. 2. Analog to Digital Converter:The attenuation rate from pass-band to stop-band is finite in any real low pass filter. If we design a signal processing system only awkwardly follow Nyquist Theorem, aliasing will be introduced into the system for the non-ideal anti-aliasing filter. In this thesis,the Oversampling Techniques are to be adopted to solute this problem. It can help us in two ways. First, it is an effective way to eliminate the aliasing. Second,it can improve the Signal to Noise Ratio (SNR) in an ADC, which equals to improve the precision of conversion. 3. Digital Filter: For the lack of prior knowledge about the environment where the robot works, a fixed filter can hardly effectively eliminate the noises. So far, the adaptive noise canceling method is the most effective one to eliminate the noise in an uncertain environment. In this thesis, a 64th order, using LMS algorithm, adaptive FIR filter are adopted and optimized for the application platform-TMS320LF2407.
修改评论