SIGNAL-TO-NOISE RATIO ESTIMATION ALGORITHM FOR SPEECH SIGNALS
Read the full article
We propose an estimation algorithm for integral value of signal-to-noise ratio (SNR) and its values in frequency domains for sound records quality testing in the speakers’ verification system. The algorithm is characterized by robustness with respect to large variability in recording conditions and sound record quality, as well as the possibility to perform SNR estimation in real time, i.e. at the rate of the incoming signal. The algorithm is based on new procedures of noise spectrum estimation and speech frames detection. The algorithm permits to evaluate SNR in the range from +6dB to +26dB for recordings with a duration of 10s or longer.