| Summary: | 碩士 === 國立成功大學 === 工程科學系 === 104 === The purpose of speech enhancement is that let the system output the target signal with the lowest noise. It can help recognize the target signal as like speech recognition. Quadcopter now is the most popular kind of drone. But applications with quadcopter are just using camera to capture the video without audio. It is necessary to reduce the noise when flying the quadcopter otherwise the target signal might be hard to be recognized.
This thesis proposes a speech enhancement system using Wiener filter based on a priori signal-to-noise ratio (SNR) to reduce additive stationary noise helping recognize the target audio signal when flying the quadcopter. First, decompose the signal into frames, and do the Fast Fourier Transform (FFT). Then, compute the coefficient of transfer function based on a priori SNR, and compensate the coefficient. Finally, do the inverse FFT to translate the spectrum and reconstruct the signal to get the non-noise signal.
In this thesis, not only run the simulation on MATLAB, but also implement the algorithm on Raspberry pi 3. It is necessary to consider the quality of the output signal and the embedded environment like Raspberry pi 3 when choosing the ideal compensatory coefficient on this speech enhancement.
Finally, verify the result by using frequency-weight segmental SNR (fwSNRseg). The speech enhancement system in this thesis can output an effective signal when the fwSNRseg of input signal is higher than -10 dB.
|
|---|
