Integrated noise reduction and acoustic echo cancellation in hands-free systems

In many hands-free voice communication applications, the presence of acoustic noise and echo degrades the quality and intelligibility of speech. This paper presents frequency-domain techniques for integrating noise reduction and echo cancellation algorithms. The self-excited method based on the short-term spectral magnitude technique is used for noise reduction. The integrated systems are implemented in frequency domain using both fast Fourier transform (FFT) and frequency-sampling filter (FSF) for splitting the time-domain signals into bandpass channels. Variable-length windows are used for voice activity detector. Computer simulations are presented to compare the performance of these techniques.